shinyreact splits a Shiny app along a clean line. The Shiny server does reactive computation. The UI is a React client that you own. shinyreact is the bridge between them, and it ships zero UI components of its own.

You can install it from CRAN or PyPI:

install.packages("shinyreact")

pip install shinyreact

shinyreact is new, and so is the way of building Shiny apps it proposes. We intend to keep the API small, but expect it to evolve as we learn from early adopters.

Shiny + React?

For most Shiny apps, defining the UI in your app.py or app.R file allows you to construct a complete, production-ready app in just a few lines of code.

The trouble starts when the design asks for something Shiny and bslib don’t have: a unique layout, richer interaction, or a component from a non-Bootstrap design system. At that point, Shiny hasn’t had the right tool for the job.

React is that tool, for three reasons:

• Ecosystem. React is the most widely used UI library on the web. Design systems, charts, tables, and maps are all one npm install away.
• The right model. React components are functions of state, which fits Shiny’s reactive model naturally. When the server sends new data, the UI re-renders efficiently.
• AI assistance. LLMs have trained on an enormous amount of React code. Ask a frontier agent for a UI and it will produce better React than it will bespoke Shiny UI. This aligns with Shiny’s goal that app developers should never be required to write low-level HTML or JavaScript themselves.

Later in the post, we’ll discuss a genomics app that renders a 584,000-cell UMAP on the GPU from a plain Shiny server. First, the basics.

Old Faithful with shinyreact

Here is the classic Old Faithful histogram app as a shinyreact app:

library(shiny)
library(shinyreact)

# Set up the page UI using shinyreact
ui <- page_react()

server <- function(input, output, session) {
  x <- faithful$waiting

  breaks <- reactive({
    seq(min(x), max(x), length.out = input$bin_count + 1)
  })

  # Use `reactive_output()` to send data to the client
  output$dist_data <- reactive_output({
    bins <- hist(x, breaks = breaks(), plot = FALSE)
    list(breaks = I(bins$breaks), counts = I(bins$counts))
  })
}

shinyApp(ui, server)

import numpy as np
import pandas as pd
from shiny.express import input
from shinyreact import reactive_output, set_react_page

# Set up the page UI using shinyreact
set_react_page()

x = pd.read_csv("faithful.csv")["waiting"].to_numpy()


@reactive_output
def dist_data():
    breaks = np.linspace(x.min(), x.max(), input.bin_count() + 1)
    counts, _ = np.histogram(x, bins=breaks)
    return {"breaks": breaks.tolist(), "counts": counts.tolist()}

Two things are different from a traditional Shiny app.

1. The UI is one line. page_react() (or set_react_page() in Shiny Express) serves the React client that lives in your app’s www/ directory. There’s no sliderInput() or plotOutput(). The www/ directory contains the static assets for the React app: ui.js and ui.css (when available).
2. The output is data. reactive_output() has no matching UI function. This is a new concept for the Shiny ecosystem! renderPlot() sends an image for plotOutput() to place. reactive_output() sends plain JSON, here the histogram’s breaks and counts. Like any render function, reactive_output() re-executes each time its reactive dependencies change. The server sends facts and React decides how to present them.

Now the client:

// src/ui.tsx (which compiles to www/ui.js)
function App() {
  const [binCount, setBinCount] = useShinyInput("bin_count", 30);
  const bins = useShinyOutputValue("dist_data", null);

  return (
    <main className="layout">
      <label htmlFor="bin_count">Number of bins:</label>
      <input
        id="bin_count"
        type="range"
        min={1}
        max={50}
        value={binCount}
        onChange={(e) => setBinCount(Number(e.target.value))}
      />
      <Histogram bins={bins} />
    </main>
  );
}

If you’ve written React before, this is an ordinary component. Histogram is whatever you like: a hand-written SVG, a charting library, or a component from your design system. The only shinyreact-specific parts are two hooks:

• useShinyInput() works like React’s useState(), except that the value is also sent to the server as input$bin_count (or input.bin_count() in Python). Calling setBinCount() updates the UI and triggers the server’s reactive graph.
• useShinyOutputValue() reads the value of a reactive_output(). When the server recomputes dist_data, the component re-renders with the new data.

Those two hooks cover the vast majority of apps.

IDs and JSON are the contract

The client and server share exactly two things: IDs and JSON values. If you write the client in TypeScript, each hook takes an optional type for its value, such as useShinyInput<number>("bin_count", 30), and your editor will then flag a mismatch that Shiny.setInputValue() never could.

Here is the full round trip for the Old Faithful app:

1. The client calls useShinyInput("bin_count", 30), which sends {"bin_count": 30} to the server.
2. The server reads input$bin_count, runs its reactive graph, and computes dist_data using reactive_output().
3. The client receives the dist_data result as {"dist_data": {"breaks": [...], "counts": [...]}}, and useShinyOutputValue("dist_data") hands it to React.

That narrow boundary is what makes a shinyreact app easy to reason about. The server doesn’t know or care how the histogram is drawn, and the client doesn’t know how the bins are computed. Each side can be reviewed, tested, and rewritten independently, whether a person or an agent wrote it.

When you need more, a few other hooks are available. useShinyOutputStatus() tells you when an output is recalculating so you can show a skeleton, and useShinyMessageHandler() receives one-off messages pushed from the server with send_message(). See the JavaScript API reference for the full list.

You don’t have to write the React yourself

A fair reaction to all of this is, “but I chose Shiny so I wouldn’t have to write JavaScript!”. That’s still the goal. What’s changed is that today’s AI agents are very good at writing React, far better than they are at writing custom Shiny UI, because there is so much more React in the world for them to learn from.

With shinyreact, your job is to own the server, which is where your data and domain logic live, and to describe and review the UI. To make that concrete, both packages ship Agent Skills:

• shinyreact-build-app scaffolds a new shinyreact app from a description.
• shinyreact-convert-app opens an existing Shiny app in a browser, describes what it does in plain English, and then rewrites the UI in React against the same server.

The day-to-day loop is familiar. You edit src/ui.tsx (or ask an agent to), the build step writes www/ui.js, and you reload the running Shiny app to see the change. The server side is unchanged: runApp() or shiny run, exactly as before.

Node.js isn’t required, since a client can be a single www/ui.js file with no build step. However, we do strongly recommend it as it gives you a proper development environment: a build step gets you TypeScript, linting, and formatting.

Keep what you already have

shinyreact doesn’t ask you to throw away the rest of the Shiny ecosystem.

• Existing outputs. renderPlotly(), render.data_frame, and other render functions work as before. Drop a <ShinyOutput id="..." /> into your React tree, and the output’s JavaScript and CSS dependencies are delivered automatically.
• Modules. ShinyModuleProvider namespaces hook IDs to match a server-side module.
• Bookmarking. URL and server bookmarking seed the initial values of useShinyInput().

Testing at every layer

Because the client and server only share IDs and JSON, each layer can be tested on its own. The server is the layer most Shiny developers care about, and it needs no browser at all. Use shiny::testServer() in R, or the new local_server pytest fixture in Shiny for Python 1.8: set inputs and assert on the JSON that comes out.

def test_histogram(local_server):
    local_server.set_inputs(bin_count=10)
    data = local_server.get_output("dist_data")
    assert len(data["counts"]) == 10

The other layers have their own tools:

• The client. Ordinary JavaScript unit tests, with whichever test runner you (or your agent) prefer.
• The wire. wire_tap() for shinytest2 and WireTap for Playwright record the JSON crossing the websocket during a browser test. That gives you an end-to-end assertion on what the client actually sent and what the server actually returned, without reaching into the rendered DOM.
• The behavior. Each example app ships a FEATURES.md: a nested list where every leaf is one checkable claim about the app, written in plain English. A person can read it as a spec, and an agent with a browser can walk it and turn each claim into a deterministic check against the running app.

The testing article covers all of them.

In the wild: Plotomics Live

Over the summer, Shiny intern Samuel Bharti built a collection of bioinformatics Shiny apps. Most of them are plain Shiny and bslib. The one that reached for shinyreact did so because the visualizations demanded it.

Plotomics Live (source, DOI) is a 26-page gallery of GPU-accelerated genomics visualizations, from oncoplots to a one-million-point Xenium spatial view and an interactive 584,000-cell UMAP. Large data skips JSON entirely and moves as compact binary typed arrays straight to the GPU. Because React owns the component, a new selection updates the data in place without re-mounting the visualization or reallocating GPU buffers.

In Samuel’s words:

R stays the analysis engine, React becomes the visualization layer, and shinyreact removes the custom JavaScript bindings, manual message passing, and serialization code that used to sit between them.

What’s next

We’re working on two directions next:

• Embedding React components in existing apps, so you can adopt shinyreact one piece at a time without porting a whole app.
• Wrapping shinyreact in your own package, so you can build a component once and ship it the way bslib ships its components.

Learn more

• Documentation: posit-dev.github.io/shinyreact
• Source and example apps: github.com/posit-dev/shinyreact
• posit::conf(2026) talk slides: Beyond Bootstrap: Building Custom Shiny UI with React

Give shinyreact a try, and please let us know what you build and what breaks.

install.packages("querychat")

pip install -U querychat

This release adds several headline features, including support for multiple tables, data-dict.yml, a full-page chat layout, support for pins, and a new /handoff command. It also builds on shinychat’s recent momentum. As a result, querychat gets chat features like history and file attachments basically for free. querychat_app() provides a quick and useful way to start chatting with data and getting bespoke ggsql visualizations, and it now uses shinychat’s page_chat() for a full chat app experience.

See the R release notes and the Python changelog for the complete list, including a few changes for existing apps if you’re upgrading.

Full-page chat layout

querychat_app() / QueryChat.app() now put the chat front and center (built on shinychat’s page_chat()), leaving more breathing room for things you create within the chat.

library(palmerpenguins)

querychat_app(penguins)

from querychat import QueryChat
from palmerpenguins import load_penguins

qc = QueryChat(load_penguins(), "penguins")
qc.app()

A view of the actual data is always accessible via the data source drawer on the right-hand side. In the case of multiple tables, you’ll see the active table¹, as well as other available tables below it.

The new page() method brings this same full-page chat layout to your own apps. Your users get the chat front and center, and you can still add custom views on other pages, in the drawer, or in the sidebar.

Learn more about building custom apps in R and Python.

qc <- QueryChat$new(penguins, "penguins")
ui <- qc$page("Penguins Explorer")

from querychat.express import QueryChat
from palmerpenguins import load_penguins

qc = QueryChat(load_penguins(), "penguins")
qc.page("Penguins Explorer")

Conversation history

Another major improvement is persistent conversation history (mostly thanks to shinychat). In addition to starting new chats and returning to previous ones, conversations now persist across page reloads and timeouts. As a result, it is now much more difficult to lose your progress.

Also, now that shinychat supports editable messages, canceling responses, file attachments, and more, querychat does too.

Multiple tables

querychat now supports multiple tables in a single chat instance. If those tables reside in a singular source, like a database, you can add them all in one fell swoop with the add_tables() method.

library(querychat)

qc <- QueryChat$new()
qc$add_tables(db, c("customers", "orders", "order_items"))

from querychat import QueryChat

qc = QueryChat()
qc.add_tables(db, ["customers", "orders", "order_items"])

querychat’s query and visualization tools handle joins across these tables, so a single question can span all of them. To write a query like the one below, the LLM first needs to know what’s in each table: column names, types, and value ranges. So it starts by fetching the schema of each table it needs (“Fetch schemas”), then generates the query with that metadata in mind.

In a custom app, the new table() method gives your server code reactive access to any table, including whatever filters the LLM has applied to it. That means you can keep building your own plots and views in Shiny, and your users can drive them just by chatting.

output$order_price <- renderPlot({
  orders_tbl <- qc$table("orders")
  # LLM can perform filter queries on $df()
  orders_df <- orders_tbl$df()
  hist(orders_df$price)
})

@render.plot
def _():
  orders_tbl = qc.table("orders")
  # LLM can perform filter queries on .df()
  orders_df = orders_tbl.df()
  plt.hist(orders_df["price"])

Provide context: data-dict

querychat does its best to gather context from the data itself. When the LLM fetches a table’s schema, it gets whatever metadata querychat can compute from the data. That’s a good start, but in practice it often isn’t enough. Column names can be cryptic, coded values need decoding, and nothing in the data says what “active customer” means to your business or how tables relate. In the example above, the LLM had to infer from column names alone that orders.customer_id points to customers.id.

A data dictionary is how you fill in what the data can’t say about itself. It’s a YAML file that follows the data-dict spec. Alongside plain-English descriptions, it has its own fields for column types, allowed values, keys, and the relationships between tables. This is now the preferred way to describe your data:

tables:
  customers:
    description: One row per customer.
    columns:
      - name: acquisition_channel
        type: enum
        values: [organic, paid_search, social, referral]
        description: How the customer first found us.
  orders:
    description: One row per order.
    columns:
      - name: customer_id
        type: number(id)
        constraints: [foreign_key]
  order_items:
    description: One row per item in an order.
    columns:
      - name: price
        type: number(quantity)
        description: Item price in USD.

relationships:
  - description: Each order belongs to one customer.
    cardinality: many-to-one
    join: orders.customer_id = customers.id
  - description: Each order has one or more items.
    cardinality: many-to-one
    join: order_items.order_id = orders.id

Pass it in as data_dict = "dictionary.yml" (R) / data_dict="dictionary.yml" (Python). When the LLM fetches a table’s schema, any column you’ve documented comes straight from your dictionary, with nothing left to infer. querychat only computes metadata from the data for the columns your dictionary doesn’t cover.

Extract insights: /handoff

Over the course of a conversation, querychat tends to produce a pile of results, some more useful than others. The useful ones deserve to live on in a reproducible artifact that doesn’t depend on the chat app.

That’s the idea behind the new /handoff slash command. It’s available in every querychat app, with no setup required. When a user types /handoff into the chat input, a wizard opens where they select the results that matter, choose an output format (e.g., Quarto, marimo, Shiny, Jupyter), and add any presentation instructions for the LLM to follow when it generates the handoff document.

When the user finishes the wizard, the handoff document’s source code streams into a code editor, where they can revise it by hand or with AI assistance. A download button then gives them a zip bundle with the handoff document, a README file, and the data sources (if they’re small enough).

Chat with pinned data

querychat can now chat with data pinned to a pins board. Pass the board and the pin name, and querychat reads the pin (parquet, CSV, JSON, RDS, and more) and uses its title, description, and tags as the starting data description:

library(pins)
library(querychat)

board <- board_connect()

querychat_app(board, "my_pin")

pip install "querychat[pins]"

import pins
from querychat import QueryChat

board = pins.board_connect()

qc = QueryChat(board, "my_pin")
qc.app()

For more control, such as setting the table name used in SQL, use the new PinSource class directly. Multiple pins, or pins mixed with ordinary data frames, also work together in one chat. Everything is materialized into a shared DuckDB connection behind the scenes, so the LLM can join and filter across all of it. See the data sources guide for R and Python for details.

A few changes for existing apps

This release also includes a handful of breaking changes, mostly around how querychat manages connections and bookmarking now that history is built in. If you’re upgrading, skim the R NEWS or Python CHANGELOG breaking-changes sections before you do.

Learn more

• querychat documentation (R) — full guides on data sources, context, tools, and deployment
• data-dict — the data dictionary spec querychat now reads
• ggsql — the grammar of graphics for SQL that powers querychat’s visualizations
• shinychat (R) — the chat UI toolkit querychat builds on
• chatlas (ellmer) — the underlying LLM tool-calling libraries
• Source on GitHub — issues, discussions, and contributions welcome

Acknowledgements

We thank everyone who contributed to these releases, for opening issues, submitting pull requests, and providing feedback: @gadenbuie, @hadley, @iainwallacebms, @iamYannC, @jnhyeon, @kolabearafk, and @thisisnic.

We are absolutely thrilled to announce the release of ggsql 0.5.0, the first beta release of ggsql since the initial release back in April. While this release brings a lot of core improvements, the beta label marks the maturity of the project more than any specific feature in this release. That being said, the features included are exciting so let’s tell you all about them.

A new reader paradigm

ggsql is modular by design with reader modules taking care of interacting with the various backends where your data live. We want ggsql to not be a monolith but instead be able to integrate itself into whatever data setup you or your organization uses, and readers is our way of making this happen.

We already showed this flexibility in the first release that included both a DuckDB, a SQLite, and a generalized ODBC reader. Since then we have added support for the new and more performant ADBC driver spec. Between this and ODBC it’s fair to say that all widely used databases are accessible, though work still remains to take full advantage of each database’s strengths.

One thing that was missing from our initial design was support for read-only database connections. ggsql heavily caches calculations on the backend using TEMP TABLE but this requires write access which you may not have. To fix this use-case ggsql 0.5.0 now includes a hybrid-reader mode. In this mode the initial data query is read from the backend database and then immediately transferred to an in-memory database of your choosing (e.g. DuckDB or SQLite) for further processing. This obviously helps in the cases where you don’t have write access to the backend, but can also speed up execution if your backend is not optimized for analytical queries. On the flip-side, it does incur an overhead when the data being plotted is huge.

You can turn on the hybrid mode by passing a cache-compatible reader to the --cache arguments in the CLI, e.g. 

ggsql exec --reader odbc://... --cache duckdb` "VISUALIZE ..."

or by prefixing it to the reader url separated by a + in the kernel:

-- @connect: duckdb+odbc://...
VISUALIZE ...

We are excited by what this new setup offers, both in allowing more users to use ggsql, but also when looking ahead and thinking about interactivity in ggsql graphics where you might rightfully not want to hit your database backend every time a user hovers over your plot somewhere.

A new writer enters the stage

While the improvements in the reader setup may be largely invisible to the user, another foundational change will surely command attention. ggsql has since its inception relied on Vega-Lite for the actual rendering. This meant that ggsql converted the query into a json spec and handed it off to the Vega-Lite JavaScript library for further processing. This choice meant that we could iterate quickly on the core of ggsql without getting bogged down by the complexity of actual rendering. From the start we knew that Vega-Lite was not meant to be the only writer in ggsql, but as development progressed it became clear that it was unsuitable in general and had to be completely replaced. Thus, I have spent my summer creating a new writer, which in time will replace Vega-Lite completely. The benefits of this are already many (and we will go through them below), but the effort will also continue to pay dividends as we are more free to support the features we deem relevant without relying on underlying support from another library.

File type galore

Outputting a JSON spec for further JavaScript processing is reasonable if you target a live HTML document but otherwise a user would generally expect an image of some sort. With the new writer we are finally able to provide this out of the box.

And we haven’t stopped at PNG…

The new writer is capable of rendering PNG, JPEG, TIFF, WebP, SVG, and PDF documents out of the box, hopefully serving all your export needs. The SVG output is optimized for further editing in vector graphics software, while the PDF output is optimized for stable rendering across machines and thus automatically embeds the glyphs used in the document.

Output type is automatically deduced from the file extension in the CLI, so you can for instance create a PDF version of a plot directly with

ggsql exec --output plot.pdf "
VISUALIZE bill_dep AS x FROM ggsql:penguins
DRAW density"

In Positron, the new formats are available as export options from the plot pane.

A native viewer

While not an output format per se, the new renderer also allows to output to a native window for previewing, directly from the CLI. This functionality can be accessed with the new view command to ggsql, e.g.

ggsql view "
VISUALIZE bill_dep AS x FROM ggsql:penguins
DRAW density"

Plotting capabilities enabled by the new writer

While the different output modes are the direct effect of a writer that can render to more formats, the new writer also provides pure, dataviz bliss by raising the ceiling for what can be plotted. This was our main reason for moving away from Vega-Lite and the reason why we can’t keep the old writer around forever — it is simply not capable enough. We will continue to reap the benefits of this in the future, but we have already included some in this release:

Rich text support through extended markdown

Text is important in data visualizations. This position should come as no surprise to those who have followed my years of trying to improve the font and text rendering capabilities of R. Just because we have moved to SQL doesn’t mean that I have given up on that quest. Thus, the new writer has full support for markdown, along with modern font support including font features and font variations. As the latter two are still not reachable from ggsql we’ll focus on the markdown support here.

By default, markdown parsing is turned on in all titles and labels, but not in break labels. Once we settle on a theming system this will all be configurable, though. Lastly, markdown parsing can also be turned on for the text layer by setting parse => true. The markdown is heavily inspired by the flavor I developed in the R marquee package. This means that everything you expect from standard markdown is available, plus a number of enhanced features:

• Italic (*), Bold (**), Underline (_), Strikethrough (~~), _Subset (~), ^Superset (^), code (`), Links ([text](url))
• Headings 1 — 6 (# — ######)
• Quote blocks (>)
• Code blocks (```)
• Bullet lists and numbered lists (- or * for former, 1. for latter)
• Horizontal lines (* * *)
• Images (![alt text](url))

From marquee it gains the support for ad-hoc coloring and sizing through the custom span syntax: {.red I'm colored red} and {#0000FF I'm colored blue} allows for named and hex-encoded colors to be set on a piece of text, and {.40 I have big letters} lets you set a font size on the fly. The customization provided by the writer goes even deeper but again, the bottleneck right now is the theming system, not the renderer. Let’s see it all in action:

VISUALIZE species AS fill, species AS x FROM ggsql:penguins
DRAW bar
SCALE fill TO ('steelblue', 'goldenrod', 'forestgreen')
LABEL 
  title => '![](logo.png) Distribution of penguin (*Pygoscelis*) species',
  subtitle => 'This plot focuses on 3 species: {.steelblue *P. adeliae*}, {.goldenrod *P. antarcticus*}, {.forestgreen *P. papua*}'

Support for captions

A small but clear deficiency of Vega-Lite was the lack of support for captions. Captions are often used to provide source information for the data, so that it travels along with the visualization:

FROM ggsql:penguins
VISUALIZE species AS fill, bill_dep AS x, bill_len AS y 
DRAW point
LABEL 
  caption => 'Data source: [Gorman KB *et al.*](www.doi.org/10.1371/journal.pone.0090081)'

A render-specific side note. If you export the above to either SVG or PDF the included link will be live and take you to the article.

Minor breaks are now supported

Visible, but ignored in the two preceding examples is the appearance of minor breaks in the output. This concept doesn’t exist in Vega-Lite and we have thus waited for a new renderer to turn it on. The interface is much like the major breaks: provide a count or an array of exact locations for the minor break to appear:

FROM ggsql:penguins
VISUALIZE species AS fill, bill_dep AS x, body_mass AS y 
DRAW point
SCALE x
  SETTING minor_breaks => 3
SCALE y
  SETTING minor_breaks => (3200, 4200, 5200)

True variable line aesthetics

The old renderer, like ggplot2, supported varying width and color along a line by chopping it up in small segments. This pragmatically works, but the only way it can look correct is to use round end caps which both forces round corners and destroys transparency due to segment overlap. In the new renderer I’ve gone out of my way to make this look correct, by converting the line into a mesh and rendering it as triangles. This means that gradient and variable width lines are now fully supported at the low level and not through any hacks:

VISUALIZE Date AS x, Temp AS y FROM ggsql:airquality
DRAW line
  MAPPING Ozone AS stroke, Wind AS linewidth

(not endorsing the above visualization in any way)

Looking forward

We have covered a lot of ground in the 5 months or so since the first release, but we have much to add still and we can’t wait. Without really going into details with any of them, I see 3 major efforts leading up to ggsql exiting beta, along with the myriad of smaller enhancements that are sure to come along:

1. Support for table output. As recently showcased in my posit::conf(2026) talk we are actively working on bringing the table formatting features from gt great-tables. This will allow you to format publication quality tables directly from your SQL query and make pure (gg)SQL driven reports an obvious option for many use cases.
2. Plot composition. One plot is good, many plots are (sometimes) better. Inspired by what the R patchwork package has done for ggplot2 plots we want to allow composition of plots, tables, and perhaps more. Who knows if one day a single SQL query can create a beautiful dashboard. The good news is that the new renderer has all of the composition logic built in already (faceting is done through that API), so much of this is a question of landing the right syntax (something we take extremely seriously).
3. Interactivity. For certain tasks, interactivity is a superior solution (though not for all). ggsql does not aim to be a D3-like playground for highly customized interactive visualization, but it does want to allow visualizations to be interactive to the extent the syntax allows without collapsing. I’m really looking forward to this design work.

These are the main focus for us, but the writer switch has itself offered so much low-hanging fruit for us to pick, that there will be plenty of user visible additions over the next year.

pip install -U shiny

The highlights: * test_server() runs your app’s server logic in memory so you can test it without a browser, * ui.page_html() lets a complete HTML document (say, the index.html your JS bundler emits) be the app’s UI, and * session.allow_reconnect() lets the browser reconnect to a live session after a dropped connection.

Full details are in the Shiny for Python changelog.

In-memory server testing

Until now, testing a Shiny for Python app meant one of two things: unit-test the pure functions your server calls, or spin up the app and a browser with Playwright and test end to end. There was nothing in between for the part that actually makes an app a Shiny app: the reactive graph.

New in v1.8, shiny.testserver.test_server() runs a server function, Express app, or shiny.App against a mock connection. There’s no browser and no network server, and your test stays a plain synchronous function: test_server() drives the event loop for you, so you never write async def or await. (If your test already runs inside an event loop, use test_server_async() instead.) Set inputs, let the reactive graph settle, and assert on outputs, all in an ordinary pytest test. It’s the Python counterpart to Shiny for R’s testServer().

For the common case, an app.py next to your test file, the new local_server pytest fixture is the whole setup:

def test_doubling_app(local_server):
    local_server.set_inputs(name="Ada", n=10)

    assert local_server.is_ok
    assert local_server.get_output("greeting") == "Hello, Ada!"
    assert local_server.get_output("doubled") == "20"

    # Inputs you don't name keep their values, so `name` is still "Ada".
    local_server.set_inputs(n=21)
    assert local_server.get_output("doubled") == "42"

local_server is the in-memory sibling of the existing local_app fixture, but function-scoped: a session remembers the inputs set so far, so each test gets a fresh one.

Each output also reports how it turned out, which is what to reach for when the assertion isn’t about equality:

def test_reports_a_bad_value(local_server):
    local_server.set_inputs(n=-1)

    assert local_server.is_ok is False
    failed = local_server.get_output("doubled")
    assert failed.status == "error"
    assert "must be positive" in failed.error

Modules work too. Reach into one with its namespaced id ("counter-n"), or take a scope and use the bare ids the module’s own code uses:

def test_counter_module(local_server):
    counter = local_server.make_scope("counter")
    counter.set_inputs(n=7)
    assert counter.get_output("label") == "n=7"

And when you need something other than app.py, call test_server() directly as a context manager. It accepts a path, a shiny.App, or a bare server function (handy for testing a module’s server function on its own):

from shiny.testserver import test_server


def test_the_other_app():
    with test_server("other_app.py") as ts:
        ts.set_inputs(n=10)
        assert ts.get_output("tripled") == "30"

The bundled shiny-for-python Agent Skill has a new test-server topic as well, so coding agents reach for in-memory server tests instead of hand-built sessions or a browser when only server logic needs checking.

New to testing Shiny apps? Start with Unit testing and End-to-end testing on the Shiny for Python website, then browse the testing API reference.

Bring your own HTML document

Shiny’s ui.page_*() functions build the HTML document for you. That’s usually what you want, but sometimes you already have one: the index.html a JS bundler like Vite emits, a hand-written template, or a page produced by another tool entirely.

ui.page_html() takes that document, as a string or a Path, and serves it as-is. Shiny’s own HTML dependencies (plus any you pass via extra_deps=) are inserted where you put a placeholder <meta> tag, and their files are served by the app:

<!doctype html>
<html>
  <head>
    <meta name="shiny-dependency-placeholder" content="">
    <script type="module" src="/assets/index.js"></script>
  </head>
  <body>
    <div id="app"></div>
  </body>
</html>

from pathlib import Path

from shiny import App, ui

app_dir = Path(__file__).parent


def server(input, output, session):
    ...


app = App(ui.page_html(app_dir / "index.html"), server)

If your document marks the spot differently, set deps_replace_pattern=. And because ui.page_html() returns a regular UI object, you can return it from a UI function (App(ui=lambda request: ...)), which is what bookmarking requires.

In Express, pass the document to ui.page_opts(html=). The whole app is routed through ui.page_html(): top-level UI markup is dropped, since the document already is the page, but any HTML dependencies it brings along are kept.

This is the Python counterpart to Shiny for R’s shinyApp(ui = htmlTemplate("index.html", document_ = TRUE)) with attachDependencies().

Session reconnection

When the websocket between the browser and the server drops, Shiny shows the “Disconnected from server” overlay and the client gives up. If your hosting environment keeps sessions alive after a client disconnects (Posit Connect and Shiny Server both can), that’s a missed opportunity: the session is still right there.

session.allow_reconnect(True) tells the client to instead show a countdown dialog and try to reconnect. On success, the browser sends its current input values back to the server, and the server recalculates outputs and sends them down again. Users pick up right where they left off after a flaky Wi-Fi blip or a laptop lid closing.

from shiny import App, ui


def server(input, output, session):
    session.allow_reconnect(True)
    ...


app = App(ui.page_fluid(...), server)

Pass "force" to attempt the reconnect anywhere, which is useful for exercising the countdown UI on a local shiny run server (where the attempt starts a fresh session rather than resuming the old one). This is the Python counterpart to Shiny for R’s session$allowReconnect().

Deprecation: ui.output_text_verbatim()

ui.output_text_verbatim() now emits a ShinyDeprecationWarning. It has long been a leftover from Shiny for R’s verbatimTextOutput(), and Python has had clearer names for both jobs for a while:

• For code or other monospaced text, use ui.output_code() with @render.code.
• For plain text, use ui.output_text() with @render.text.

The matching Playwright controller, playwright.controller.OutputTextVerbatim, is deprecated alongside it; use playwright.controller.OutputCode instead.

Other improvements

A few more changes worth a quick mention. The full list is in the changelog:

• Closing a session no longer destroys the reactive values and calcs created in it. Since v1.6.1, refreshing the page while an @reactive.extended_task was in flight could raise DestroyedReactiveError once it settled. Values and calcs are now left readable at their last value and reclaimed by garbage collection; effects are still destroyed on close.
• ui.input_slider() and ui.update_slider() now encode datetime.date values as UTC midnight, matching Shiny for R, so they no longer shift by a day when the server runs in a timezone ahead of UTC. Naive datetime.datetime values round-trip unchanged too.
• @render.download_button and @render.download_link now honor @output(id=). Previously the URL and the registered handler disagreed on the id, so clicking the control returned a 404.
• @expressify and @render.express no longer fail with RuntimeError: Failed to find function '...' in AST when another decorator has changed the function’s __name__, a pattern often used to give each @render.express function in a loop a unique output id.
• Navsets created with an id now use it as their data-tabsetid, so tab panes get stable DOM ids instead of ones built from a random integer. This makes the markup reproducible and easier to target from custom CSS and JavaScript. (Thanks, @pevolution-ahmed!)
• ui.show_offcanvas() now accepts the id of an ui.offcanvas() panel already in the UI, matching ui.hide_offcanvas() and ui.toggle_offcanvas(). It also accepts bare tag content, wrapping it in a new anonymous panel.
• @render.data_frame now renders data frames whose column names are empty or not strings; column ids are positional and never derived from the name.
• @render.ui outputs inside a ui.popover() or ui.tooltip() without a title= no longer get stuck showing “recalculating”.
• shiny run --app-dir <dir> <app> now honors --app-dir for Shiny Express apps.
• ui.input_task_button(type=None) no longer drops the class its input binding needs, so the button is bound and clicking it works.
• playwright.controller.Offcanvas gains open(), loc_trigger, loc_title, loc_footer, and expect_title(), expect_footer(), and expect_placement().
• playwright.controller.InputSelectize no longer clicks the page body to close its dropdown, so a test can’t accidentally fire the app’s own click handlers.

In closing

We’re excited to see what you build (and test) with this release. As always, if you have questions or feedback, join us on Discord or open an issue on posit-dev/py-shiny. Happy Shiny-ing!

Acknowledgements

A big thank you to all the folks who helped make this release happen by opening issues and contributing code:

@ambevill, @arabidopsis, @bealdav, @chernojagne, @ChidiebereNjoku, @cpsievert, @danieldebondt-tf, @drewe7192, @eeshsaxena, @ErdaradunGaztea, @FBruzzesi, @gadenbuie, @jat255, @jubilee2, @karangattu, @kramerrs, @MichielNoback, @mykolaskrynnyk, @nightcityblade, @nvelden, @pevolution-ahmed, @schloerke, @shawnboltz, @tjpalanca, @weichisyu, and @xiruizhao.

Last week, three packages in Posit’s open-source AI stack shipped significant releases. We introduced commons 0.1.0, and ellmer and shinychat both received substantial updates. These releases are part of a broader effort to make it as easy as possible to build modern chat applications in R and Python.

In this newsletter, we’ll take a quick tour of what’s new.

Introducing commons

commons, a new framework for building trustworthy self-service data analysis agents in R and Python, is now on CRAN. Read the full announcement here.

The Python package is currently in a pre-release beta stage, with more features arriving over the next few weeks.

If you’re a data analyst, data scientist, statistical programmer, or other data practitioner, you likely have extensive domain knowledge and a collection of trusted code that you already use in analyses, apps, reports, and packages. The core idea behind commons is that we can leverage this trusted code to improve an agent’s correctness.

A commons agent first searches for a trusted calculation. If it finds one that can answer the user’s question, it can run that vetted code and the answer is deterministically marked as verified.

If it doesn’t find a trusted calculation, the agent searches trusted context before writing custom R, Python, or SQL. The answer is either given a citation or marked as “untrusted,” depending on whether the agent provides a verified citation that supports its approach.

The model doesn’t decide how trustworthy its answer is. commons assigns each label deterministically based on the analysis path taken by the agent.

commons also ships with an agent skill to help you create a commons agent and functions for analyzing your users’ conversations.

shinychat v0.5.0 (R) and v0.7.1 (Python)

shinychat v0.5.0 for R and v0.7.1 for Python bring together more of what you need to build a complete chat application. Several of these shinychat updates also made commons possible!

Read the full blog post here. There are many more updates worth checking out.

page_chat()

Use page_chat() instead of the bslib page_*() functions when you want the chat to be the center of your application. page_chat() creates a full-window, chatbot-oriented layout with support for navigation pages, conversation history, an artifact drawer, and more.

Conversation history

Conversation history is enabled by default when you use chat_server() in R or Chat(client=...) in Python, allowing users to start a new conversation, switch between saved conversations, search them, rename them, and delete them.

Readable tool calls and citations

This shinychat release also includes several improvements for understanding how a model arrived at its response.

One such improvement is readable tool calls. By default, related tool calls are grouped into compact, single-line “activity rows”, keeping them from overwhelming the conversation. You can still inspect the individual tool calls by expanding a row.

shinychat also displays citations returned by providers’ built-in web-search and web-fetch tools.

ellmer 0.5.0

ellmer 0.5.0 is now on CRAN. ellmer makes it easy to work with LLMs from R.

Read the full announcement here. Many of the features made available in this release are also available in recent releases of chatlas, ellmer’s sibling package in Python.

Citations

When a model uses a supported built-in web tool, ellmer now returns and displays the provider-supplied citations. This works with claude_tool_web_search(), claude_tool_web_fetch(), google_tool_web_search(), and openai_tool_web_search(), helping you identify the sources behind the model’s answer.

chat <- chat_openai()
chat$register_tool(openai_tool_web_search())
chat$chat(
  "What is the most recent version of ellmer on CRAN? Look it up."
)
#> The most recent CRAN release of **ellmer** is **version 0.5.0**, published
#> **September 4, 2026**.
#> ([cran.r-project.org](https://cran.r-project.org/package%3Dellmer))[1]
#>
#> Sources
#> [1] CRAN: Package ellmer: https://cran.r-project.org/package%3Dellmer

Tokens and costs

Managing tokens and costs is an important part of working with LLMs.

Ever want to know how many tokens an input will take before sending it? For supported providers, you can now use chat$token_count() to estimate input token use. Instead of actually sending the request to the model, it sends the request to the provider’s token-counting API.

chat <- chat_openai(model = "gpt-5.6-luna")
prompt <- content_pdf_file("example-document.pdf")
chat$token_count(prompt)

#> [1] 252

This estimates only the tokens used by the input and does not predict the number of output tokens, so it won’t represent the total round-trip count.

Companies frequently release new models and change their prices. Use the new function models_update_prices() to download and cache the latest pricing data from the ellmer GitHub repo. Cost estimates reported by token_usage(), Chat$get_cost(), Chat$get_tokens(), and printed Chat objects use this data.

Send files to the model

It’s often useful to send files as part of a chat. You can now send CSV, Markdown, code, and other text-based files to a model with content_document_file() and content_document_url(). For large files or files reused across multiple turns, if you’re using chat_openai(), chat_anthropic(), or chat_google_gemini(), use chat$file_upload() instead. It uploads the file once and returns a reference for $chat(). This avoids repeatedly sending the file and reduces token usage and cost.

Solid improvements for custom agents

Taken together, these releases make it easier to build more complete custom agents. ellmer manages model interactions in R, shinychat provides the user-facing chat interface, and commons adds a framework for data analysis agents that builds upon these two.

The package is built on ellmer, chatlas, and shinychat, Posit’s open source LLM stack. You can use whatever model you want from any of the providers supported by those packages with it.

To install the R package, run:

install.packages("commons")

To install the Python package from PyPI, run:

pip install commons

The Python package is currently in a pre-release beta stage, but you can install and play around with it today, with more features arriving over the next few weeks.

Design philosophy

AI agents for data analysis can range from overly cautious and narrowly correct to wildly and confidently incorrect. commons provides a framework for you to design more trustworthy analysis agents by providing them with access to existing trusted code, while still allowing them enough flexibility to answer novel, realistic questions.

If you are a data analyst, data scientist, statistical programmer, or other data practitioner, you likely have a deep understanding of your problem domain and a collection of trusted code you depend on for your analyses and use to create apps, reports, and packages. The core idea behind commons is that we can improve an agent’s correctness by giving it the right access and documentation to run this code that you have already vetted.

When answering questions, commons agents first search through a pool of trusted code. If the agent finds an appropriate piece of trusted code, it can invoke it directly, and its response will be tagged with a green shield icon . If it doesn’t, it will search through relevant context before writing its own SQL, R, or Python. If the agent can find trusted context that justifies its approach, it can provide a citation to it at the end of its answer, which will be deterministically checked by commons. Otherwise, the answer is marked with a small warning label .

Notably, the agent itself does not decide how to label a response. commons instead labels answers deterministically, based on the path the agent takes to get it to its answer.

Get started

To get started with the R package, check out the introductory vignette. The package ships with an agent skill to help you hook your trusted code and context up to the agent.

The Python package (although still in beta) is based on the same ideas, and the resulting apps will look very similar regardless of whether you use R or Python—they literally share the same CSS! Check out the Python package site to learn more.

shinychat is a toolkit for building complete, conversation-centered chat applications with Shiny. The R package pairs with ellmer, and the Python package pairs with chatlas. Install the latest releases from CRAN or PyPI:

install.packages("shinychat")

pip install -U shinychat

We cover a lot in this post, and there’s even more in the releases. See the R release notes and the Python changelog for the complete list of changes, including a few changes for existing apps if you’re upgrading.

Build a chat application

A useful chat application needs more than a text box and a streaming response. Your users need a way to return to an earlier conversation, start a new one, correct a question, compare answers, inspect sources, and see what the model is doing when it calls a tool. They may also need to upload a file, open a preview, or move between the chat and the rest of the application.

shinychat gives you sensible starting points for building that experience. Pair it with ellmer in R or chatlas in Python, and you can get a working chat app running with little setup. The chat application model has three layers:

1. page_chat() gives you a full-window chat app with space for navigation, history, tools, and supporting content.
2. chat_ui() lets you place chat wherever it fits best in your application.
3. chat_server() for R or Chat(client=...) for Python connects your app to an ellmer or chatlas client and enables the integrated chat features.

When you want a fully custom experience or need a model client other than ellmer or chatlas, the lower-level pieces are still available for you to assemble yourself.

Start with page_chat()

When chat is the center of your application, use page_chat(). It gives your users a full-window experience with a chat home, navigation pages, sidebars, toolbars, conversation history, and an artifact drawer. Users can move to a settings or sources page while their conversation keeps working and streaming. The Get started guide for R and the Page chat guide for Python walk through the full layout.

Create a chat app

Build a shinychat application starts similarly in both languages:

library(shiny)
library(shinychat)

ui <- page_chat(title = "Assistant", id = "chat")

server <- function(input, output, session) {
  client <- ellmer::chat_openai(
    system_prompt = "You are a helpful assistant."
  )

  chat_server("chat", client)
}

shinyApp(ui, server)

from chatlas import ChatAnthropic
from shinychat.express import Chat, page_chat

client = ChatAnthropic(system_prompt="You are a helpful assistant.")
chat = Chat(id="chat", client=client)

page_chat(title="Assistant", id="chat")

With just a few lines of code, you’ll have a working chat app backed by a live LLM. Passing a client to chat_server() in R, or to Chat() in Python, does all the hard work for you, fulling connecting your app to the model client and giving you a complete multi-user chat application¹.

For a personal chat UI you can use while you develop locally, pass an ellmer client to chat_app() in R, or a chatlas client to Chat(client=...), and then call .app() in Python.

Welcome users

When you’re app opens, don’t leave your users hanging with an empty chat canvas, gree them with chat_greeting() (R, Python)!

Greetings can be used to explain the application, set expectations, and give users a useful first step before they write their first message. By default, they disappear when the user starts chatting, but you can set persistent = TRUE in R or persistent=True in Python to keep one at the top of the conversation history.

Greetings are written in markdown and can even provide actionable suggestions. Users can click a suggestion to fill the input, ready to edit before sending, or send it immediately.

## Welcome!

What would you like to do?

* <span class="suggestion submit">Summarize my data</span>
* <span class="suggestion">Create a plot</span>
* <span class="suggestion">Explain this code</span>

You don’t have to greet your users with the same message every time, you can use LLMs to generate fresh custom greetings. To learn more, we’ll point you to the chat_greeting() documentation pages (R, Python), but it’s worth noting that dynamic greetings can stream into the chat like any other response.

Return to earlier conversations

One of the biggest features to arrive in this release is conversation history, giving your chat app the ability to save and return to previous conversations. It will also persist the current conversation across page reloads and other disconnects, virtually eliminating the possibility of losing your progress. As usual, when you connect shinychat with an ellmer or chatlas client, conversation history is wired up and enabled for you!

Save conversations

The history drawer lets users:

• Start a new conversation.
• Switch between saved conversations.
• Search conversations.
• Rename a conversation.
• Delete a conversation.
• Return to the conversation that was active when they last opened the app.

shinychat generates a short title once the conversation has enough content. Users can replace that title, and title generation never overwrites a manual rename.

• Rename
• Search

You can history_options() in R or HistoryOptions in Python to configure the conversations that shinychat saves. The main options are:

• restore_mode, which controls which conversation opens when a user returns to the app:
  • "browser" is the default. It returns that browser to its most recent conversation without changing the URL.
  • "url" puts the active conversation ID in the address bar, so users can bookmark or share a specific conversation.
  • "bookmark" restores the conversation with the rest of the app state when your app uses Shiny server bookmarking.
• store controls where shinychat saves conversations. Use "memory" for local development or tests, or "file" to save them on disk.
• title controls how the automated conversation titles are generated.

For example, this configuration stores conversations on disk and puts the active conversation ID in the URL:

history <- history_options(
  restore_mode = "url",
  store = "file"
)

chat_server("chat", client, history = history)

from shinychat import Chat
from shinychat.types import HistoryOptions

history = HistoryOptions(
    restore_mode="url",
    store="file",
)

chat = Chat("chat", client=client, history=history)

On Posit Connect, conversation history is included with the platform and is enabled automatically when you provide a model client. The default configuration uses Connect’s persistent storage and scopes conversations to the authenticated user. That gives every user a private conversation history without an additional history service or per-user setup.

In every restore mode, shinychat keeps the transcript in its configured store instead of putting the full conversation in the URL.

Edit a message and compare answers

Editing a message now creates a new conversation branch. When a user edits and resends an earlier message, shinychat forks the conversation at that point: the original question and its later messages remain on one branch, while the edited question begins another. Users can move between the answers with the branch controls in the message.

• Branch 1
• Branch 2

Branches help when a prompt is almost right or when a model takes an unhelpful direction, and they make comparing answers easy without starting over. And they are part of the saved conversation, so users return to their place in the conversation after a reload.

Add content and controls

When chat is part of a larger application, your users still need access to filters, settings, sources, and results. page_chat() gives you a place to put those alongside the conversation: a drawer for results, toolbars for controls, and offcanvas panels for settings you would rather keep off screen.

Artifact drawer

chat_drawer() gives you a place to show previews, rendered reports, tables, plots, or other bits of Shiny UI next to your chat. Your users can keep the conversation visible while they inspect a result.

See the drawer documentation for R or Python for the full API. The complete application example combines a drawer with the rest of the application layout.

Toolbars

Your app may need a Help button that works on every page, while the chat home needs an action that’s only relevant when you’re looking at the conversation. page_chat() gives each action a home through scoped toolbars, built on the toolbar components that bslib and Shiny shipped earlier this year.

If you want an action to follow users through the whole app — pass it to toolbar_global. Put chat-home actions in toolbar in page_chat(), and give a secondary page its own toolbar through chat_nav_panel(). toolbar_input puts related actions below the message box.

See the R get started guide or the Python Page chat guide for the full toolbar API.

Offcanvas panels

page_chat() pairs offcanvas panels with secondary content, such as an answer-length slider or citation setting, and a toolbar button can open an Answer settings panel from any page:

Complete application

As your app grows, page_chat() can grow around the conversation. You can add secondary pages and a sidebar for filters or other app UI and the application menu keeps those options available on narrow screens.

The following example brings the toolbars, sidebar, navigation, and drawer together.

ui <- page_chat(
  "Research assistant",
  id = "chat",
  toolbar = bslib::toolbar(
    bslib::toolbar_input_button(
      "clear_chat",
      "Clear conversation",
      icon = bsicons::bs_icon("arrow-counterclockwise")
    )
  ),
  toolbar_global = bslib::toolbar(
    bslib::toolbar_input_button(
      "help",
      "Help",
      icon = bsicons::bs_icon("question-circle")
    )
  ),
  sidebar = chat_sidebar(
    tags$p("Use filters to focus the results."),
    history = FALSE
  ),
  pages_navbar = list(
    chat_nav_panel(
      "Sources",
      tags$p("Sources selected during this session appear here."),
      toolbar = bslib::toolbar(
        bslib::toolbar_input_button(
          "refresh_sources",
          "Refresh",
          icon = bsicons::bs_icon("arrow-repeat")
        )
      )
    )
  ),
  drawer = chat_drawer(
    tags$p("Select a result to inspect it here."),
    title = "Latest result",
    open = FALSE
  )
)

from faicons import icon_svg
from shiny import ui
from shinychat import chat_drawer, chat_nav_panel, chat_sidebar
from shinychat.express import page_chat

page_chat(
    "Research assistant",
    id="chat",
    toolbar=ui.toolbar(
        ui.toolbar_input_button(
            id="clear_chat",
            label="Clear conversation",
            icon=icon_svg("arrow-counterclockwise"),
        )
    ),
    toolbar_global=ui.toolbar(
        ui.toolbar_input_button(
            id="help",
            label="Help",
            icon=icon_svg("question-circle"),
        )
    ),
    sidebar=chat_sidebar(
        ui.p("Use filters to focus the results."),
        history=False,
    ),
    pages_navbar=[
        chat_nav_panel(
            "Sources",
            ui.p("Sources selected during this session appear here."),
            toolbar=ui.toolbar(
                ui.toolbar_input_button(
                    id="refresh_sources",
                    label="Refresh",
                    icon=icon_svg("arrow-repeat"),
                )
            ),
        )
    ],
    drawer=chat_drawer(
        ui.p("Select a result to inspect it here."),
        title="Latest result",
        open=False,
    ),
)

Show how the model reached an answer

Understanding how an LLM arrived at an answer is just as — if not more — important than getting the answer from the model. A response can include ordinary text, thinking content, web activity, citations, tool calls, tool results, and custom UI. shinychat works hard to make the model’s work visible and presents each part in a way that helps users understand the answer and what produced it.

Keep tool calls readable

Tool calls are now shown as compact activity rows instead of letting them take over the conversation, refining the tool-call cards shinychat introduced last year. By default, related calls are grouped together into a single row, and users can still expand a group, open an individual call, and inspect the request and result when they need more detail.

Opening an individual call shows the request and the result in a card:

Grouping keeps the answer readable, and the request and result stay one click away. To customize grouping or register tools, see Tool UI in shinychat for R, Tools in Shiny for Python, tool/function calling in ellmer, or tool calling in chatlas.

Show citations for web search and fetch

Many LLM providers offer built-in web search and web fetch tools that let your agent search the web, and their APIs return citations when the model uses that content in a reply. shinychat now displays those citations automatically.

For example, here’s how to register Claude’s tools with an ellmer or chatlas client:

library(ellmer)

client <- chat_anthropic()
client$register_tool(claude_tool_web_search())
client$register_tool(claude_tool_web_fetch())

from chatlas import ChatAnthropic, tool_web_fetch, tool_web_search

client = ChatAnthropic(
    kwargs={
        "default_headers": {
            "anthropic-beta": "web-fetch-2025-09-10"
        }
    }
)
client.register_tool(tool_web_search())
client.register_tool(tool_web_fetch())

When this client is used with chat_server(), citations are connected directly to the portions of the assistant’s response that they support.

Custom retrieval applications, like the RAG systems you can build with ragnar or raghilda, can use the same citation UI by prompting the assistant to use a <shiny-aside> tag to attach a source to a claim.

Stream responses and show thinking

With chat_server() in R or Chat(client=...) in Python, shinychat streams responses and shows supported thinking content in a collapsible panel. Users can cancel a slow response with the stop button or the Escape key, and the partial response stays in the conversation.

Add files and shortcuts

Attach files

File attachments are now supported in shinychat! Your users can send images, PDFs, and text files through a file picker, drag and drop, or paste, and shinychat sends each file to the model alongside the user’s message. When you use chat_server() in R or Chat(client=...) in Python, your app gets that support for free.

Add slash commands

You can now register chat shortcuts, or slash commands, with chat$slash_command() in R or @chat.slash_command() in Python. The command palette appears when users type /, and they serve as a way to trigger server-side code, inject context or additional prompting, or even just take an action in your app, all from the chat input.

Check out the shinychat for R or shinychat for Python documentation for details.

More shinychat-powered apps

The next release of querychat will bring these chat features to data applications, including conversation history, attachments, tool displays, and citations. It will introduce a page-first querychat_app() workflow and a new page() API for adding querychat to an existing Shiny page.

btw 1.5.0 already uses shinychat 0.5.0 to give btw_app() a complete coding assistant for your R projects. It adds conversation history, a page_chat() layout, and slash commands to an assistant that can use your R session, project files, and package documentation.

A few changes for existing apps

Existing chat_ui() applications remain supported when chat shares a page with other top-level content. When the conversation should fill the application instead, choose page_chat() and use it as the outermost page container; nesting it inside another page layout breaks the full-window layout and history experience.

In R, chat_mod_ui() and chat_mod_server() are soft-deprecated in favor of pairing chat_ui() and chat_server() by ID. In both languages, a startup message no longer seeds a conversation when history is enabled; use a greeting or append messages through the chat object instead.

The release also protects users from unsafe model-authored Markdown, shows an error when a response fails before streaming starts, and preserves tool results, citations, attachments, and other rich content when users return to a conversation.

With page_chat(), chat_server() or Chat(client=...), and the history options, you can now give your users a complete chat application: saved conversations they can return to, messages they can edit into new branches, greetings and suggestions to start from, and responses with visible tool calls, citations, and thinking.

Read the shinychat for R documentation or the shinychat for Python documentation to explore the examples. For the complete list of changes, see the R release notes and the Python changelog.

Acknowledgements

We thank everyone who contributed to these releases, for opening issues, submitting pull requests, and providing feedback: @bastianolea, @bianchenhao, @christophsax, @cpsievert, @crissthiandi, @elnelson575, @gadenbuie, @Harshit28j, @JamesHWade, @jcheng5, @jlxAtNovozymes, @jnhyeon, @jose-c-milliman, @kaipingyang, @lucasrod16, @markmcd, @nbenn, @parmsam, @schloerke, @shea-parkes, @simonpcouch, @slupczynskim, @thisisnic, @wlandau, and @xx02al.

You can install the latest version from CRAN with

install.packages("ellmer")

This blog post covers the major changes in this release: a lifecycle update, updates to how you can work with files with ellmer, new ways to update model price data, returning citations when using web search tools, and new hooks for developers building on ellmer’s tool loop.

The full list of changes can be found in the release notes.

Lifecycle

chat_github() and models_github() are now defunct, since GitHub Models has been retired.

We’ve tightened up what a tool can return: a string, an atomic vector, a JSON string, or a Content object. Returning anything else, like a data frame or a list, now gives a deprecation warning. Previously ellmer converted these to JSON for you, but any problem with the conversion surfaced long after your function had finished, and it was easy to forget that the model can only read the result, not compute with it. For a data frame, convert it yourself:

get_weather <- tool(
  function(cities) {
    df <- weather_api(cities)
    jsonlite::toJSON(df, dataframe = "columns")
  },
  ...
)

New features

Sending files to the model

There are now two ways to give the model a file. New content_document_file() and content_document_url() send text-based documents like CSV, Markdown, and code files, just as content_pdf_file() and content_image_file() already do for PDFs and images. The contents go inline with your message, so this works with every provider:

penguins <- tempfile(fileext = ".csv")
readr::write_csv(
  data.frame(
    penguin = c(
      "Waddlesworth", "Flipper McGee", "Turbo Tuxedo",
      "Captain Blubber", "Sir Slidesalot"
    ),
    race_time_seconds = c(43.2, 38.7, 31.9, 45.1, 36.4)
  ),
  penguins
)

chat <- chat_anthropic()
chat$chat("Who won the race?", content_document_file(penguins))
#> Based on the race times, **Turbo Tuxedo** won the race with the fastest time of
#> **31.9 seconds**.

Inline contents are re-sent with every turn, which adds up over a long conversation, especially with a large file. For those cases, chat$file_upload() sends the file to the provider once and returns a reference you pass to $chat() instead. Because the file isn’t re-sent with every message, this also reduces your token usage and costs:

chat <- chat_google_gemini()
race <- chat$file_upload(penguins)
chat$chat("Who won the race?", race)
#> **Turbo Tuxedo** won the race with the fastest time of **31.9 seconds**.
chat$chat("And who came last?")
#> **Captain Blubber** came in last with the slowest time of **45.1 seconds**.

You can manage your uploads with chat$file_list(), $file_get(), $file_download(), and $file_delete(). File management works with chat_openai(), chat_anthropic(), and chat_google_gemini(), and replaces the now-deprecated claude_file_upload() and google_upload().

Citations

When a model answers using a built-in web search or fetch tool, the provider usually reports which sources back the answer. ellmer 0.5.0 captures citations from Claude, Google, and OpenAI.

chat <- chat_anthropic()
chat$register_tool(claude_tool_web_search())
chat$chat("What are the current stable versions of R and Python? Look them up, one line each, no commentary.")
#> R: R version 4.6.1 (Happy Hop) has been released on 2026-06-24[1]
#>
#> Python: Python 3.14.7 / 5 August 2026[2]
#>
#> Sources
#> [1] R: The R Project for Statistical Computing: https://www.r-project.org/
#> [2] Python (programming language):
#> https://en.wikipedia.org/wiki/Python_(programming_language)

Citations are also kept in the chat history and included in streamed output, so apps built on ellmer can display them as well.

Counting tokens and keeping prices current

Two additions make it easier to know what a conversation will cost before you commit to it. Chat$token_count() asks the provider how many tokens some input would use, without actually sending it to the model.

chat <- chat_anthropic()
prompt <- "Tell me a joke about an R programmer"
chat$token_count(prompt)
#> [1] 19

Token counting is currently supported by chat_anthropic(), chat_openai(), chat_google_gemini(), chat_google_vertex(), and chat_posit().

Providers change their prices more often than we release ellmer. You can now update in between releases with models_update_prices(), which downloads the latest pricing data from GitHub and caches it locally.

Other improvements

• Default models have been updated across providers: chat_anthropic(), chat_aws_bedrock(), chat_databricks(), chat_posit(), and chat_snowflake() now use Claude Sonnet 5; chat_openai() and chat_openrouter() use GPT 5.6 Terra; and chat_google_gemini() and chat_google_vertex() use Gemini 3.7 Flash. We update the default models regularly, so if you’d prefer to pin your code to a specific model, you should specify it using the model parameter.
• chat_aws_bedrock() now supports Bedrock Mantle, the newer endpoint that serves models like Claude Mythos and the GPT-5 family through the Anthropic Messages and OpenAI Responses APIs, rather than only the Converse API. ellmer picks the right API from the model name, so this should just work. If you’re using a model it doesn’t recognize, you can set the new api argument yourself.
• You can now stream structured output. Chat$stream() and $stream_async() gain a type argument, which works the same way as in $chat_structured(), for providers that support it.

Developer updates

Two new features help developers building on ellmer’s tool loop.

Inside a tool, tool_context() returns the request that triggered it and the conversation so far, so a tool can make decisions, not just the model. Here, a query tool stops after three calls:

run_query <- tool(
  function(sql) {
    # count_tool_results() is a stand-in for your own helper
    if (count_tool_results(tool_context()$turns) >= 3) {
      tool_reject("Query budget used up. Answer with what you have.")
    }
    jsonlite::toJSON(DBI::dbGetQuery(con, sql), dataframe = "columns")
  },
  name = "run_query",
  description = "Run a SQL query",
  arguments = list(sql = type_string())
)

Chat also gains $on_request_start() and $on_request_end(), which fire before and after every request to the model, including each round of the tool loop. For example, to time each request:

chat <- chat_anthropic()
chat$register_tool(run_query)

started <- NULL
chat$on_request_start(\(turns) started <<- Sys.time())
chat$on_request_end(\(turn) message("Request took ", round(Sys.time() - started, 1), "s"))

chat$chat("Find the mean of every column in mtcars, one query at a time.")
#> Request took 3.6s
#> ◯ [tool call] run_query(sql = "SELECT * FROM mtcars LIMIT 5")
#> ● #> [{"mpg":21,"cyl":6,"disp":160,"hp":110,"drat":3.9,"wt":2.62,"qsec":16.46,…
#> Now I'll compute the mean of each column one query at a time, as requested.
#> Request took 2.7s
#> ◯ [tool call] run_query(sql = "SELECT AVG(mpg) AS mean_mpg FROM mtcars")
#> ● #> [{"mean_mpg":20.0906}]
#> Request took 1.9s
#> ◯ [tool call] run_query(sql = "SELECT AVG(cyl) AS mean_cyl FROM mtcars")
#> ● #> [{"mean_cyl":6.1875}]
#> Request took 2s
#> ◯ [tool call] run_query(sql = "SELECT AVG(disp) AS mean_disp FROM mtcars")
#> ■ #> Error: Tool call rejected. Query budget used up. Answer with what you
#> have.
#> It looks like the query budget has been used up, so I can only report the means
#> I was able to compute before being cut off:
#>
#> | Column | Mean |
#> |--------|------|
#> | mpg | 20.0906 |
#> | cyl | 6.1875 |
#> Request took 3.4s

$on_request_start() also receives the turns about to be sent, so an agent can compact its history with chat$set_turns() before the context window fills up.

Acknowledgements

A big thanks to the 73 people who helped make this release possible by filing issues, contributing code, and asking questions: @1beb, @abiyug, @aclink88, @AdaemmerP, @alesanGreat, @Apollo7777777, @arnavchauhan7, @arunrajes, @atheriel, @awunderground, @bakaburg1, @bastianolea, @bshor, @cerebrixos, @CoryMcCartan, @cpsievert, @D-M4rk, @dareneiri, @debruine, @diegomsg, @diegoperoni, @dipterix, @earthcli, @etiennebacher, @feddelegrand7, @FrancescoMonti-source, @frankiethull, @gadenbuie, @hadley, @hectorgray, @hopessugar, @hswerdfe, @JamesHWade, @jamesinottawa, @jcheng5, @jcrodriguez1989, @jeroenjanssens, @JosiahParry, @jrosell, @kaipingyang, @karawoo, @kbenoit, @kchou496, @klin333, @kolabearafk, @ksr-zguo, @lazasaurus-ai, @lionel-, @MLiedgens, @n8layman, @nbenn, @neil-bray, @nrineausanofi, @ntentes, @omorante, @petzi53, @rajabzadehalidip, @rempsyc, @Sade154, @sarahsdao, @scjohannes, @simonpcouch, @Sirhubi007, @SokolovAnatoliy, @sounkou-bioinfo, @stefanlinner, @t-kalinowski, @Tazinho, @thisisnic, @thoov08, @trangdata, @WvdH-Novus3, and @xmarquez.

This post is about the R package. There is also a Python version of orbital that works on scikit-learn pipelines.

You can install both from CRAN with:

install.packages(c("orbital", "tidypredict"))

This post covers the highlights of these releases. You can see the full list of changes in the orbital release notes and the tidypredict release notes.

A lot more models

Previously orbital and tidypredict supported a handful of useful models. This included linear models, decision trees, random forests, and boosted trees models. This release adds discriminant analysis, naive Bayes, neural networks, support vector machines, partial least squares, and several more rule and ensemble methods. This fills in the gaps of most of the known models that can be made to work with orbital.

Newly supported for regression:

• bart(engine = "dbarts")
• boost_tree(engine = "h2o_gbm")
• linear_reg(engine = "glm")
• mlp(engine = "nnet")
• null_model()
• pls(engine = "mixOmics")
• rand_forest(engine = "aorsf")
• rand_forest(engine = "partykit")
• rule_fit(engine = "h2o")
• svm_linear(engine = "kernlab")
• svm_linear(engine = "LiblineaR")

Newly supported for classification:

• bag_tree(engine = "rpart") and bag_tree(engine = "C5.0")
• boost_tree(engine = "C5.0")
• boost_tree(engine = "h2o_gbm")
• C5_rules(engine = "C5.0")
• decision_tree(engine = "C5.0")
• discrim_linear(engine = "MASS"), discrim_quad(engine = "MASS")
• discrim_linear() with the "mda", "sda", and "sparsediscrim" engines
• logistic_reg(engine = "LiblineaR")
• mlp(engine = "nnet")
• multinom_reg(engine = "nnet")
• naive_Bayes(engine = "klaR") and naive_Bayes(engine = "naivebayes")
• null_model()
• pls(engine = "mixOmics")
• rule_fit(engine = "h2o")
• rule_fit(engine = "xrf")
• svm_linear(engine = "kernlab")
• svm_linear(engine = "LiblineaR")

One thing worth noting for models with “h2o” in their engine value, is that you need a running H2O cluster to build the orbital object. Prediction and SQL generation from the orbital object stays dependency free.

Here is an example of linear discriminant analysis model, which was not supported before this release, fit on the penguins data and then run in DuckDB.

library(tidymodels)
library(discrim)
library(orbital)
library(duckdb)

data(penguins, package = "modeldata")
penguins <- tidyr::drop_na(penguins)

rec_spec <- recipe(species ~ ., data = penguins) |>
  step_dummy(all_nominal_predictors()) |>
  step_normalize(all_numeric_predictors())

mod_spec <- discrim_linear(engine = "MASS")

wf_spec <- workflow(rec_spec, mod_spec) |>
  fit(penguins)

orb <- orbital(wf_spec, type = c("class", "prob"))

con <- DBI::dbConnect(duckdb::duckdb())
penguins_db <- dplyr::copy_to(con, penguins, "penguins")

predict(orb, penguins_db)

# A query:  ?? x 4
# Database: DuckDB 1.5.5 [root@Darwin 25.6.0:R 4.6.1/:memory:]
   .pred_class .pred_Adelie .pred_Chinstrap .pred_Gentoo
   <chr>              <dbl>           <dbl>        <dbl>
 1 Adelie             1.000        6.43e- 9     4.64e-25
 2 Adelie             1.000        5.25e- 6     8.81e-16
 3 Adelie             1.000        1.97e- 4     2.61e-17
 4 Adelie             1.000        2.99e- 8     1.93e-23
 5 Adelie             1.000        8.34e- 9     4.42e-29
 6 Adelie             1.000        1.60e- 6     4.14e-20
 7 Adelie             1.000        6.03e-10     1.45e-18
 8 Adelie             0.999        6.47e- 4     4.16e-20
 9 Adelie             1.000        9.56e-10     1.97e-30
10 Adelie             1.000        6.73e-14     9.13e-27
# ℹ more rows

We have documented the list of supported models. If there is a model you need that is still missing, let us know so we can prioritize it.

When a model has no probability

While we added these new methods we ran into a problem we didn’t have before. For many classification models you get both predicted probabilities and hard class predictions. This release has added models where that is not the case. A number of newly added models only produce hard class predictions.

Instead of trying to invent class probabilities, we produce an informative error for the affected models.

c5_wf <- workflow(
  species ~ .,
  decision_tree(mode = "classification", engine = "C5.0")
) |>
  fit(penguins)

orbital(c5_wf, type = "prob")

Error in `orbital()`:
! "prob" predictions are not available for this model.
ℹ It predicts a class directly, with no probability behind it.
ℹ Use `type = "class"` instead.

tidypredict is now a toolkit, not just a function

We have talked a lot about orbital so far. This is because we think that it is the ideal interface compared to tidypredict, if your goal is to generate SQL expressions for a fitted model or workflow.

orbital and tidypredict work together to generate the expression that you need. We have expanded tidypredict with a number of generics, which are all developer focused. With the goal that packages other than orbital can benefit from the work we have done. Three of them describe what a model’s fitted expressions compute:

• tidypredict_output_type() returns "numeric", "prob", "decision", or "class"
• tidypredict_outcome_levels() returns the outcome levels in model order
• tidypredict_normalized() reports whether per-level probabilities already sum to one

Below we see two of the generics in action.

library(tidypredict)

svm_fit <- svm_linear(mode = "classification", engine = "LiblineaR") |>
  fit(sex ~ bill_length_mm + body_mass_g, data = penguins)

tidypredict_output_type(svm_fit$fit)

[1] "decision"

tidypredict_outcome_levels(svm_fit$fit)

[1] "female" "male"  

The other five expose the per-tree pieces that tidypredict_fit() assembles, so a package generating its own code can split an ensemble apart and put it back together:

• tidypredict_trees() returns per-tree expressions
• tidypredict_n_trees() returns the number of trees
• tidypredict_combine_trees() turns per-tree expressions back into a prediction
• tidypredict_class_trees() returns per-tree expressions for each outcome level
• tidypredict_class_exprs() returns one finished expression per outcome level

rf <- rand_forest(mode = "regression", trees = 5, engine = "ranger") |>
  fit(body_mass_g ~ bill_length_mm + flipper_length_mm, data = penguins)

tidypredict_n_trees(rf$fit)

[1] 5

tidypredict_combine_trees(rf$fit, rlang::syms(paste0("tree_", 1:5)))

(tree_1 + tree_2 + tree_3 + tree_4 + tree_5)/5

A lot of this logic was hardcoded inside orbital on a model by model basis, it is now formalized in such a way that you can easily use just the bits you need.

Acknowledgements

Many thanks to all the people who contributed to orbital and tidypredict since the last release!

@EmilHvitfeldt, @jannikbx, and @RAMitchell.

Welcome back to another edition of our monthly Positron updates! Each month we share highlights from our latest release and useful resources. Last release we told you about new Data Connections sources, a round of polish for inline output in Quarto documents, and help installing missing packages. This milestone brings a redesigned welcome page, a first version of Import Data, new Posit Assistant features, expanded Data Connections, package vulnerability scanning, and a more responsive Console.

Welcome page refresh and interpreter setup

We redesigned the Positron welcome page. It now leads with an environment setup card that checks whether Python and R are ready to use, with actions to help resolve any problem it finds. The Positron badge and name now come with a Help button that opens the Help pane, and a banner links to the walkthroughs, including a new “Get Started with Positron” walkthrough that covers the Positron panes, keyboard shortcuts, built-in extensions, and Git.

The environment setup theme continues into interpreter selection itself. When you select a Python managed by your operating system or a package manager, Positron now offers to create a virtual environment for your workspace instead of installing packages into that shared interpreter. With no folder open, the environment Positron creates now lives at ~/.virtualenvs/positron rather than ~/.venv, so it stays in the interpreter picker after a restart, and Positron asks before creating an environment in your home directory. We also removed the confusing startup notification that reported no interpreters were found while linking to documentation saying no setup was needed, and interpreters installed under /opt/python are now labeled Global instead of Unknown in the interpreter picker.

Import Data

Before we started work this month, importing data via the Positron UI was our most upvoted feature request and this release now delivers a first version. When you view a CSV or TSV file in the Data Explorer, a new Import Data button in the action bar opens a dialog. The dialog shows the code to load that file into a data frame. You can copy the code, or click Import to run it in the console, starting a session if one is not already running.

Import Data supports CSV and TSV files in Python with pandas and in R with the readr package, as well as Excel workbooks and Parquet files (with readxl and nanoparquet in R). The generated code can reproduce the filters and sorts you have applied in the Data Explorer, and it names the file by a workspace-relative path when the file is inside your workspace, so the code is easier to share and rerun. You can open the dialog from the Data Explorer, the File menu, the Variables pane, or the File Explorer context menu.

Posit Assistant

This release brings a new Agent Layout that opens Posit Assistant in the editor area with a compact Session pane, giving you more visibility into the agent’s actions as it works alongside your code. Configuring language model providers also gets a redesign. Our new Configure LLM Providers modal groups providers by connection state, so you can see at a glance which providers are ready to use. If you have trouble with the new dialog and need to switch back, set assistant.newProviderModal to false.

You can now configure multiple custom providers, each with its own name, type, endpoint, credential, and model list; the previous single “Custom Provider” option is now called “OpenAI Compatible” to better match what it actually does. Amazon Bedrock users get a smoother experience as well. An expired AWS SSO session can be renewed right from the provider modal instead of requiring aws sso login in a terminal, and the AWS profile and region can now be set in the configuration dialog rather than only through environment variables or a hand-edited providers.json. Speaking of which, providers.json now accepts comments, and your comments survive edits that Positron makes to the file.

Data Connections

The Data Connections preview keeps growing. A new ODBC data connection driver lets you browse any database with an installed ODBC driver in the Connections pane and open it in the Data Explorer. Data sources already configured on your machine appear automatically. Databricks gains OAuth sign-in on desktop and reads DATABRICKS_TOKEN, DATABRICKS_HOST, and DATABRICKS_CONFIG_FILE credentials managed by Posit Workbench automatically. A Disconnect option in the context menu closes a connection and any Data Explorers opened from it, and each data connection driver now gets its own log output channel.

Smaller improvements round out the preview. You can now set dataConnections.enabled per workspace, so a repository can turn on the Connections pane for anyone who opens it. The tree is shallower so table and column names get more of the panel’s width, and a new dataConnections.tree.indent setting controls the indentation. DuckDB connections now default to read-only, so the Connections pane and a Python or R session can have the same database open at once; when a lock conflict does happen, the error now explains that another session has locked the database.

Package security vulnerabilities

The Packages pane now shows known security vulnerabilities (Common Vulnerabilities and Exposures scoring) for installed Python and R packages, so you can see at a glance whether something in your environment has a known CVE. The data comes from your environment’s own Posit Package Manager repository when it has one, and from the public Posit instance otherwise, so what you see reflects the same package source your organization already governs.

A more responsive Console

Console code submission is now faster, always shows visual feedback, and can be canceled while a completeness check is in flight; the Console no longer waits indefinitely with no feedback when a kernel is slow or unreachable. The Console breaks multi-statement input into complete expressions and executes them statement by statement for languages that support it. A new console.promptWhenIncomplete setting runs submitted code immediately without a completeness check.

A few more Console fixes are worth knowing about. The Console no longer takes focus at runtime startup when you are working in another view or editor, Interrupt stays visible after switching between busy consoles, and session names now ellipsize to fit as the console tab list narrows.

What’s coming next

• posit::conf(2026) is next week! Our team will have several sessions on Positron, and there is still time to register to join us virtually from anywhere in the world.
• Meet Posit at CDAO Government on September 22-23 in Washington, D.C. Stop by our booth to talk data modernization and where agentic AI fits in government.

Recently, there’s been a lot of talk about fine-tuning models. The reasoning usually goes something like this:

• There’s some task that needs to be done regularly.
• Today’s frontier models can do it quite reliably, but it’s expensive and the bills are starting to rack up.
• Cheaper models can’t quite do the task reliably, but a fine-tuned variant of a small, open-weights model might be able to just as cheaply.
• Therefore, you should fine-tune models for tasks that your organization does.

I’ve seen a number of blog posts cited in support of this idea.

It makes sense that fine-tuning is appealing! AI is getting expensive, many players in the space cannot make the privacy guarantees we’d hope, and just as you start to rely on one proprietary model, it’s phased out in favor of a newer release. That said, my reaction is that this approach seems more engineering-intensive and, likely, more expensive than alternative approaches. I’ll try to make that case here.

What is fine-tuning?

Before we go further, it’s probably worth quickly outlining what I mean by fine-tuning. In short, fine-tuning means training an existing model further to improve its performance on a particular task.

Large language models contain huge arrays of parameters. Input, usually in the form of text, is turned into arrays of numbers, and those arrays get multiplied a bunch of times to form the output, which is then typically converted back into text. Those multiplications are quite computationally intensive, and the array of parameters is itself quite large. For example, the smallest models that are beginning to be able to reliably complete basic agentic work contain 8 billion parameters (in short, “8B”).

Fine-tuning changes a model’s behavior, either by updating the existing parameters, or by training a smaller set of additional parameters that work alongside the originals. The goal of fine-tuning is, broadly, to increase the performance of a model on a given task while minimizing the impact on the model’s ability to do other tasks.

Fine-tuning is hard

It is very, very difficult to successfully fine-tune a model. I don’t claim that it’s impossible, but it is a substantial engineering effort, requiring the careful attention of dedicated scientists with access to today’s frontier models across several weeks, just for the first edition of the model. To show why, let’s revisit each step of that process.

What model should you start with? In short, you want the smallest possible model that has the ability to learn to do the task reliably while retaining sufficient general intelligence. It is very hard to predict when or how capabilities will emerge during fine-tuning without first fine-tuning, meaning that the choice will need to be revisited several times once you’ve made a first go at the remaining steps.

What data gets used for training? Once you’ve chosen a model, you need to find some training data to fine-tune it with. What should you use? Ideally, you would use the real data that represents the task and its inputs. However, in order to train on that real data, you typically need the user’s consent, and it’s likely that you won’t have that consent.

And then, when you do have user consent and choose to train on that data, you now have a mandate not to overfit. This is because if the model you’re fine-tuning internalizes the data you’re training on and is able to (even hazily) recollect it, you’ve now exposed that data to any users of the model.

The other approach, then, is synthetic data, created by asking a model to generate a bunch of scenarios that resemble the real inputs and then demonstrate how to carry out the task in those scenarios. But this type of synthetic data has its own problems: notably, the models creating the data have their own set of tics that make the data unrepresentative of real-world data. The Elias in the Lighthouse effect is a notable example of this phenomenon. A wide variety of models will, when asked to write a story, write about a lighthouse keeper named Elias Thorne. On its own, the story of Elias Thorne might be informative training data. However, 1,000 stories about a lighthouse keeper Elias Thorne are not.

How do you nudge the weights? Let’s assume you do have a diverse, representative set of training data to work with. You’ll now need to decide how, mechanistically, to change the behavior of the model—do you need a full fine-tune, or will a LoRA be sufficient?

Next, you’ll need to tune a set of 5-10 hyperparameters. Notably, because they are hyperparameters, there is no generally good “magic number” for them, and instead you’ll need to try a bunch of values and see what works. You’d make some guesses, see what happens, then generate some hypotheses on what a given change to those parameters might do, then try out a different number and see if it has the desired effect. Every time you’re figuring out what to try next, you’ll be reading a ton of test cases and trying to observe general failure modes exhibited in this intermediate draft of your fine-tuned model. That might change the initial choice of model you’re fine-tuning, and it might change which subsets of the training data you’re exposing to the training process.

How do you measure success? You need a reliable way to tell whether each version of your model is actually improving. That is especially tricky for free-text tasks, because two answers can be equally good even when they differ syntactically. Because of this, you’ll probably want an LLM-as-a-judge system, where another model compares the responses to a target and grades according to a rubric you’ve supplied. LLM-as-a-judge systems are themselves quite hard to design correctly, and you’ll be reading a bunch of that judge’s grading transcripts, too.

It’s also worth mentioning that getting your fine-tuned model to score well on your own benchmark is not the hard part. Many of these fine-tuning writeups include a plot along these lines:

When I see that a single-digit-billion-parameter fine-tuned model scores better than Fable 5.1 or GPT 5.6 Sol on an evaluation, I interpret that as evidence that the evaluation is not meaningful. In my own experience, it is not (comparatively) hard to get a small model to score very well on any given benchmark. What’s much more difficult is preserving the broad intelligence of a model while driving the evaluation score up. A meaningful evaluation can do both at once, measuring task performance under a realistic, broad distribution of possible task configurations. It is very hard to author meaningful evaluations, especially for models as capable as those that exist today.

But let’s say you’ve made it this far! You found a good model to start from, a diverse set of training data that you’ve obtained consent to train on, and a meaningful way to measure progress. Now, it’s time to put it in production.

Fine-tuning is expensive

Fine-tuning can seem appealing because the actual fine-tuning part of the process can be relatively inexpensive, likely on the order of a few dollars to a few hundred dollars.¹

However, actually deploying the model to do the task you trained it to do is likely to be substantially more expensive.

…because hosting is expensive

Hosting a model yourself is more expensive than using a similarly capable model hosted by someone else unless you have extraordinary volume. Frontier-model providers like Anthropic and OpenAI serve extraordinarily large volumes, keeping their compute almost always near max capacity. It’s likely that the same won’t be true for your fine-tuned, self-hosted model.

As an example, let’s say I host Gemma 4 26B A4B on a single H100. I can rent a high-availability H100 GPU instance at $0.0833 a minute, or $44,000 a year. Assuming a coding-agent-like workload,² that would buy about 96 billion Sonnet 5 tokens, and Sonnet 5 is a much more capable model. To break even, this fine-tuned model would need to serve about 182,000 tokens per minute, 24/7, throughout the year.

It might be possible to serve that many tokens. The hard part is finding enough demand for one task to keep the model busy, and that demand needs to be relatively smooth. If the model sits idle, you’re still paying for it. If demand spikes, you either rent a second GPU or accept worse performance for users.

…and you must host

The other suggestion here is that, if the model is small enough, users who need to do the task can just download the weights and run it on their laptop or a dedicated workstation.

The effectiveness of this argument depends on the kind of task to be done. However, it’s worth considering the potential for this to be a very unpleasant user experience compared to just using a model that someone else hosts. For example, let’s say my colleague does some task for an hour every week and there’s some model small enough to run on my laptop that’s capable of doing the task. In order to use the model for that task, my colleague would need to download the 4GB or 8GB or 100GB or whatever of weights. That user would also need to be capable of configuring the serving of the model, and they’d need to make that happen every time they started the task. If that model was accessed through a tool that the user was using regularly anyway—Claude Code pointed at a local ollama model, for instance—they’d need to remember to switch the model over in the application’s settings. If that model was accessed through a different interface, the user would need to use a different interface than they normally use LLMs with for that task specifically. Neither of these are good UX.

Even if the software for locally serving models got much more pleasant than it currently is, it’s hard to compete with the experience of pay-as-you-go for the Everything Tool that wraps the Everything Model.

Your fine-tune will quickly fall behind

Let’s say you successfully fine-tune a model based on some fictional model Kuen 3. Then, the following week, Kuen 3.5 is released. 3 of the 100 most capable LLM scientists on this planet worked on it. It’s almost as good as your fine-tune on that specific task, and it’s also broadly capable at a very broad array of tasks.

Fine-tuning is not a boat that is lifted by the rising tide of broader AI progress—in order to take advantage of the Kuen 3.5 release with your fine-tune, you’d need to restart that process of choosing parameters and observing fine-tuning runs. The underlying architecture of the model may have changed, and thus the approaches that you used to fine-tune Kuen 3 might not work for Kuen 3.5 on your first try.

What you should do instead

Instead, the boat that is lifted by a rising tide in this context is plain old prompt engineering. (POPE, as Joe Cheng calls it.) Choose a model that’s available to you, that fits your price point, and seems broadly capable across a wide variety of public benchmarks. (Extra points if the vibes on the model from people you trust are good, and extra points if someone else is serving it.)

Then, put together a short prompt telling the model how to do the task, perhaps inside of some coding agent harness like Claude Code (or, hey, Posit Assistant), and see what it does. Adjust the prompt to tell it how to do things correctly that it tends to trip up on, and iterate from there. If you’ve already put together an evaluation as part of your fine-tuning process, you could even reuse that! My colleague Sara and I have written about prompt engineering in the past if you’re interested in learning more, once about the more specific task of POPE for teaching LLMs about R packages but in relatively generalizable ways, and once focused on deciding between the popular ways to deliver context to coding agents.

Recent past newsletters

• How to choose a model
• Keep track of your data exploration with Posit Assistant’s EDA log

Subscribe via RSS

To install the newest release, run the following in R:

install.packages("vitals")

The package includes two new helpers, claude_code() and codex(), which allow you to compare your own ellmer-built agents with leading coding agents. This release also ships another new helper, vitals_log_read(), which supports reading log files back into tibbles, including columns of resumable ellmer Chats. Finally, the release includes several performance improvements; log files are much smaller, and the log viewer that reads them is now substantively faster.

To read the full list of changes, see the changelog.

Agent solvers

vitals is a port of Inspect, a well-adopted Python framework for LLM eval from Posit’s own JJ Allaire. One of the concepts that vitals borrows from Inspect is the concept of a “solver,” or the LLM-powered system that sets out to solve some task. The simplest solver is just the LLM itself, with no system prompt or tools, like what you’d get from running chat_anthropic() from ellmer. Solvers can gain all sorts of prompts and tools, which allows vitals users to test the effect of a change in their prompt or the addition of a new tool.

In the last year or so, the dominant interface to solvers in Inspect has become “agent solvers”: interfaces to the popular coding agents Claude Code and Codex. You call the helper claude_code() or codex(), and Inspect will proxy traffic through the real coding agent harness.

vitals now has first-class support for these two helpers, allowing users to compare their own agents built with ellmer to popular coding agents like Claude Code and Codex. You provide the set of tasks and grading guidance, and vitals will take care of the communication with Inspect.

Read eval logs back into ellmer Chats

One of the big annoyances I’ve had in my own usage of vitals is log storage. So that users can use Inspect’s log viewer directly, we write evaluation logs to a JSON format that Inspect can read.^[1] However, I often want to write R code against the original R objects—ellmer Chats especially—that the logs were generated from. Loading in the ellmer Chats would especially be helpful for inspecting (ha!) the conversation histories in the same interface that users of the ellmer application would see.

Because of this, I’ve often saved both the JSON logs and .rda logs, the latter of which contain the ellmer Chats. These files are large on their own, and it feels even more silly passing around duplicates of them.

The new release of vitals introduces vitals_log_read(), which reads an eval log file back into a tibble of samples. (It’s almost exactly what you’d get if you ran the get_samples() method on a vitals Task object.) That tibble includes reconstructed solver (and, for model-graded scorers, scorer) chats as ellmer Chat objects. For some providers, the chats will even be resumable; you can load in a solver from a JSON file into an R session and ask that solver a question yourself.

Performance improvements

The long and the short of this section is just to say that:

1. Logs will take up less storage space than they did before. Roughly, logs written with the new vitals version will be 4x smaller than before, and the magnitude of savings increases with the complexity of the log.
2. We now display logs (with vitals_view()) much more quickly. The log viewer should feel very snappy for almost all uses of the package.

I’m really excited to have this release on CRAN! Take it for a spin and let me know if you run into issues on the package repository.

Welcome to our newsletter, posit::glimpse()!

If you’re currently reading this on our blog, consider subscribing to Product Updates - Open Source on our subscription page to receive this newsletter directly in your inbox.

Welcome to this month’s roundup of the latest open-source developments from Posit! Our update list is slightly shorter this time around, because our team is hard at work putting the final touches on posit::conf!

In this special feature, we are spotlighting a few presentations that will be broadcast live throughout the conference. Be sure to register today if you’d like to tune in live. And, if you enjoy being chronically online like me, I’ll be hanging out in the Discord community server throughout the event, alongside your 1000+ closest data pals 👾

Key product updates and new releases

Positron August Release Highlights

The Positron 2026.08 Release blog post highlights several key updates and improvements to Posit’s next-generation data science IDE, including expanded Data Connections (preview), more polished Quarto inline output, centralized AI provider configuration, and performance and reliability upgrades.

• Read more in the Positron August Release Highlights blog post.

Kimi K3 and GLM 5.2 are now in Posit AI

Posit has added two new open-weights models, Kimi K3 and GLM 5.2, to the Posit AI platform. Both models stream tokens nearly twice as fast as Anthropic’s models in internal testing and offer lower cost-per-token rates (partly due to more efficient tokenizers and no extra charge for cache writes).

• Read more in the Kimi K3 and GLM 5.2 are now in Posit AI blog post.
• Learn how to pick your models in the AI Newsletter: How to choose a model blog post.

cuda.ml 0.4.0

cuda.ml is an R package that brings NVIDIA GPU-accelerated machine learning directly to R workflows. This release introduces streamlined setup and developer experience and expanded tree-ensemble inference.

• Read more in the cuda.ml 0.4.0: GPU-accelerated machine learning from R blog post.

Orbital 0.6.0

Orbital converts Scikit-learn pipelines into SQL queries so that they can run in your database. Orbital 0.6.0 adds PyTorch neural network support, enabling trained torch.nn.Sequential models to compile directly to SQL for database-native inference without requiring a Python runtime, ONNX Runtime, or separate model server.

• Read more in the Neural networks in Orbital for Python 0.6.0: PyTorch straight to your database blog post.

recipes 1.4.0

recipes lets you create a pipeable sequence of feature engineering steps. Recipes 1.4.0 introduces a new way to look inside a recipe partway through, a substantial speedup for steps that are applied to many columns, and multi-column support in step_regex() and step_count().

• Read more in the recipes 1.4.0 blog post.

themis 1.1.0

themis contains extra steps for the recipes package for dealing with unbalanced data. Version 1.1.0 adds eleven new sampling steps for handling unbalanced data. It also adds setting sampling targets per class and more distance metrics.

• Read more in the themis 1.1.0 blog post.

New cheatsheets

We’ve refreshed our cheatsheets page, which now features three brand-new additions!

• Python Polars: Developed in partnership with Polars, Inc., this cheatsheet introduces Polars core concepts and key expressions.
• tidymodels: Updated cheatsheets detailing practical workflows across the entire tidymodels ecosystem.
• yardstick: Guidance on using yardstick within tidymodels to evaluate model predictive performance.

Check them below:

Python Polars: The Definitive Cheatsheet

Quick reference guide for transforming, analyzing, and visualizing data with Python Polars

Jeroen Janssens, Thijs Nieuwdorp

Great Tables plotnine

Machine learning with tidymodels

A map of the tidymodels packages, grouped by where each one fits in the machine learning workflow

Edgar Ruiz

tidymodels

Measure model performance with yardstick

Quick reference guide for measuring how well a model predicts with yardstick

Edgar Ruiz

yardstick

posit::conf(2026) talk sneak peeks

With over 100 talks, it’s impossible to feature all of them in this newsletter (but you can explore the complete lineup on the event schedule!). I wanted to give a quick preview of topics you can look forward to (whether you tune in live during the conference or catch up on demand).

1. Positron in prime time

It’s been about two years since the public release of Positron, and 🎵this IDE is on fire 🎵

The lineup of conf talks proves it. From Quarto notebooks in Positron (did you know inline output is available?!), to using Positron in education, to how it treats SQL as a first-class language, to delightful workflows in Python and R, come see what people are cooking up with Positron.

2. What can’t you do with Quarto?

The best way of discovering what Quarto can do is seeing how others are using it!

We’re starting off strong, with the first session of Virtual Day (September 14th) being “Quarto, R + Python”, where Björn Fisseler will discuss creating accessible reports and Wasim Lorgat sharing the history of notebooks. The Quarto theme continues on the 15th and 16th with talks on designing influential and scalable reports, creating learning ecosystems, teaching bilingual data science, developing wiki chatbots, and developing unreasonably effective dashboards and slides — all in Quarto.

3. I’ll definitely be joining…

   • Positron and the Three Bears: Teaching Python with Positron (presented by Marc Dotson) - Practical lessons from an educator navigating Python instruction and finding the sweet spot with Positron
   • First Impressions Matter: Styling Data Products That People Actually Want to Use (presented by Shelby Level) - Shelby creates visually striking data products, and I can’t wait to pick up some design inspiration and techniques from her session
   • A Grammar of Graphics for SQL (presented by Thomas Lin Pedersen) - We’ll hear the developer of ggsql share the story of bringing ggplot2-style visualization concepts to SQL
   • Outgrowing Spreadsheets: Rebuilding Research Data Collection with Shiny and Posit Connect (presented by Kelsey Chalmers) - Kelsey shares the before-and-after architecture and how their research team transformed a fragile, spreadsheet-based data collection process into a sustainable system using Shiny and Posit Connect

Of course, this is just the tip of the conf iceberg. One of my favorite things about conf is the diversity of talks, from real-world data science implementation, to insights from the field, to creative and whimsical experiments. I hope to see you there!

If you can’t wait til conf (or if you’re reading after conf is over!), there are other events that you can join us at:

• Data Science Lab, every Tuesday at 12pm ET
• Data Science Hangout, every Thursday at 12pm ET
• Building Repeatable AI Workflows with Custom MCP Servers, September 30
• Our next conference appearance

Have a lovely September, and reach out anytime! isabella [dot] velasquez [at] posit [dot] co

This release brings 3 main improvements that are worth your attention: a new way to look inside a recipe partway through, a substantial speedup for steps that are applied to many columns, and multi-column support in step_regex() and step_count().

You can install it with:

install.packages("recipes")

Seeing a recipe partway through

A recipe is constructed using a series of pipes. Since it is prepped as a unit, it can be a little hard to look into the internals and see what happens at each stage of the recipe. If you wanted to know what the data looked like after the second recipe step you would need to write out a separate recipe with just those steps.

This is no longer the case as bake() now takes a stop_at argument, which applies only the steps up to and including the one you specify. Below we write out a small recipe on the ames data set.

data(ames, package = "modeldata")

ames <- ames |>
  select(Sale_Price, Neighborhood, Bldg_Type, Year_Built, Gr_Liv_Area) |>
  mutate(Sale_Price = log10(Sale_Price))

ames_rec <- recipe(Sale_Price ~ ., data = ames) |>
  step_other(Neighborhood, threshold = 0.05) |>
  step_dummy(all_nominal_predictors()) |>
  step_normalize(all_numeric_predictors()) |>
  prep()

The three steps each do a job that the next one depends on. Neighborhood has 29 levels, many of them rare, so step_other() pools every level accounting for less than 5% of the data into a single "other" level, which keeps the next step from producing near-empty indicator columns. step_dummy() then turns the remaining factors (Neighborhood and Bldg_Type) into 0/1 indicator columns, since most models need numeric predictors. Finally, step_normalize() puts every predictor on the same scale (mean 0, standard deviation 1), which matters for regularized and distance-based models.

Now let’s first see the data before we apply the recipe on it.

tail(ames) |> 
  glimpse()

Rows: 6
Columns: 5
$ Sale_Price   <dbl> 5.117271, 5.153815, 5.117271, 5.120574, 5.230449, 5.274158
$ Neighborhood <fct> Mitchell, Mitchell, Mitchell, Mitchell, Mitchell, Mitchell
$ Bldg_Type    <fct> OneFam, OneFam, OneFam, OneFam, OneFam, OneFam
$ Year_Built   <int> 1960, 1984, 1983, 1992, 1974, 1993
$ Gr_Liv_Area  <int> 1224, 1003, 902, 970, 1389, 2000

If we now wanted to see what happened after the first step, we pass in stop_at = 1 to bake().

bake(ames_rec, new_data = tail(ames), stop_at = 1) |>
  glimpse()

Rows: 6
Columns: 5
$ Neighborhood <fct> other, other, other, other, other, other
$ Bldg_Type    <fct> OneFam, OneFam, OneFam, OneFam, OneFam, OneFam
$ Year_Built   <int> 1960, 1984, 1983, 1992, 1974, 1993
$ Gr_Liv_Area  <int> 1224, 1003, 902, 970, 1389, 2000
$ Sale_Price   <dbl> 5.117271, 5.153815, 5.117271, 5.120574, 5.230449, 5.274158

And we see that Mitchell was collapsed into "other" by step_other(). Next we set stop_at = 2 to see the result of step_other() followed by step_dummy().

bake(ames_rec, new_data = tail(ames), stop_at = 2) |>
  glimpse()

Rows: 6
Columns: 15
$ Year_Built                      <int> 1960, 1984, 1983, 1992, 1974, 1993
$ Gr_Liv_Area                     <int> 1224, 1003, 902, 970, 1389, 2000
$ Sale_Price                      <dbl> 5.117271, 5.153815, 5.117271, 5.120574, 5.…
$ Neighborhood_College_Creek      <dbl> 0, 0, 0, 0, 0, 0
$ Neighborhood_Old_Town           <dbl> 0, 0, 0, 0, 0, 0
$ Neighborhood_Edwards            <dbl> 0, 0, 0, 0, 0, 0
$ Neighborhood_Somerset           <dbl> 0, 0, 0, 0, 0, 0
$ Neighborhood_Northridge_Heights <dbl> 0, 0, 0, 0, 0, 0
$ Neighborhood_Gilbert            <dbl> 0, 0, 0, 0, 0, 0
$ Neighborhood_Sawyer             <dbl> 0, 0, 0, 0, 0, 0
$ Neighborhood_other              <dbl> 1, 1, 1, 1, 1, 1
$ Bldg_Type_TwoFmCon              <dbl> 0, 0, 0, 0, 0, 0
$ Bldg_Type_Duplex                <dbl> 0, 0, 0, 0, 0, 0
$ Bldg_Type_Twnhs                 <dbl> 0, 0, 0, 0, 0, 0
$ Bldg_Type_TwnhsE                <dbl> 0, 0, 0, 0, 0, 0

And we can see the dummies created on Neighborhood and Bldg_Type.

Much faster on wide data

Many of the steps such as step_normalize(), step_unknown(), and step_impute_mean() work by modifying the data frame in place. No new columns are created and no columns are removed. This type of step is quite common throughout recipes. It was found that these steps could have their performance improved, especially if applied to a lot of columns. Regardless of how many columns you have, you will get the same results as before, at the same speed or faster depending on the number of columns.

library(recipes)

set.seed(1)
df <- as.data.frame(matrix(rnorm(200 * 5000), nrow = 200))

rec <- recipe(~ ., data = df) |>
  step_normalize(all_numeric_predictors()) |>
  step_YeoJohnson(all_numeric_predictors())

system.time(prepped <- prep(rec))
system.time(bake(prepped, new_data = df))

On my machine I get roughly these numbers.

prep() got about 3× faster here and bake() over 20×, and the gap widens as you add columns.

Over 40 steps benefited from this change. See the changelog for the complete list. You don’t have to change any code to get this. Existing recipes are just faster.

If you write your own steps, the helper behind this is exported as recipes_map_cols().

step_count() and step_regex() take multiple columns

The two steps step_count() and step_regex() were always the odd ones out as they could not be applied to more than one column. This has been fixed in this release, where the steps can now be applied to any number of columns, putting them in line with the rest of the steps in the tidymodels ecosystem.

ice_cream <- tibble(
  flavor = c("rocky road", "stony brook", "vanilla"),
  notes  = c("a rock in there", "no match", "rock and stone")
)

recipe(~ ., data = ice_cream) |>
  step_count(flavor, notes, pattern = "rock", result = "rocks") |>
  prep() |>
  bake(new_data = NULL)

# A tibble: 3 × 4
  flavor      notes           flavor_rocks notes_rocks
  <fct>       <fct>                  <int>       <int>
1 rocky road  a rock in there            1           1
2 stony brook no match                   0           0
3 vanilla     rock and stone             0           1

When more than one column is selected, the new columns are named {column}_{result} rather than just {result}, so they stay distinguishable.

Acknowledgements

Many thanks to all the people who contributed to recipes since the last release!

@EmilHvitfeldt, @instantkaffee, @LeonidasZhak, and @topepo.

cuda.ml is for data scientists who work primarily in R and want to use a GPU without moving their modeling workflow to Python or learning low-level GPU APIs. Version 0.4.0 is a substantial update to the package. It makes installation much simpler, expands tidymodels support, adds more ways to run tree-ensemble models, and makes it straightforward to save and restore supported fitted models.

Install from CRAN

For most users, setup is two commands:

install.packages("cuda.ml")
cuda.ml::cuda_ml_install()

The package from CRAN is a regular, portable R package. cuda_ml_install() downloads and verifies the matching compiled backend and GPU libraries, then keeps them in a cache for later R sessions. On a supported system, you do not need to compile cuda.ml from source, configure a Python environment, or assemble the GPU libraries yourself.

The result is a familiar R package workflow: install the package, prepare its supporting libraries once, and start an analysis. The extra installation call is explicit because the GPU libraries are much larger than the R package. Repeated calls reuse the completed cache, and loading cuda.ml itself is quiet and does not initialize CUDA.

Prebuilt support is available for Linux x86_64 with glibc 2.28 or newer. GPU operations require a supported NVIDIA GPU and driver 580 or newer. On Windows, install and run R inside a compatible WSL2 Linux distribution. See the installation guide for the complete system requirements and source-build options.

Use familiar modeling interfaces

cuda.ml registers parsnip engines for linear, logistic, and multinomial regression, random forests, nearest neighbors, and radial, polynomial, and linear support-vector machines. For example, this fits a random-forest classifier on the GPU using the standard parsnip interface:

library(cuda.ml)
library(parsnip)

forest_spec <- rand_forest() |>
  set_mode("classification") |>
  set_engine("cuda.ml")

forest_fit <- forest_spec |>
  fit(class ~ ., data = modeldata::hpc_data)
predict(forest_fit, modeldata::hpc_data[1:5, ], type = "prob")

# A tibble: 5 × 4
  .pred_VF .pred_F .pred_M  .pred_L
     <dbl>   <dbl>   <dbl>    <dbl>
1    0.309  0.597  0.0666  0.0275
2    0.899  0.0838 0.0138  0.00335
3    0.965  0.0261 0.00850 0.000883
4    0.973  0.0228 0.00416 0.000352
5    0.966  0.0298 0.00416 0.000352

set_engine("cuda.ml") selects the GPU-backed cuda.ml engine; the rest is a standard parsnip workflow. Recipes can learn preprocessing on the training data and carry it into resampling and prediction.

Work directly with cuda.ml

cuda.ml also provides a direct R interface. This is useful when you prefer a function-oriented workflow or want to use the package on its own.

library(cuda.ml)
library(ggplot2)

clusters <- cuda_ml_kmeans(scale(faithful), k = 2)
faithful$cluster <- factor(clusters$labels)

ggplot(faithful, aes(eruptions, waiting, color = cluster)) +
  geom_point(size = 2.5) +
  labs(
    x = "Eruption duration (minutes)",
    y = "Waiting time (minutes)"
  ) +
  theme_minimal()

The direct API covers supervised models as well as clustering and dimensionality reduction, including DBSCAN, k-means, PCA, tSVD, UMAP, and t-SNE. It also includes stochastic-gradient-descent regression, the hyperbolic-tangent SVM kernel, and external tree-ensemble inference.

Run tree ensembles on a GPU or CPU

This release expands where and how you can make predictions with tree ensembles. The new nvForest support powers prediction for random forests trained with cuda_ml_rand_forest() and can load trained XGBoost models, LightGBM text models, and Treelite checkpoints. Once a model is loaded, the API provides standard prediction along with model information, leaf identifiers, individual-tree predictions, and checkpoint import and export.

GPU inference uses the complete cuda.ml installation. A deployment that only needs CPU inference can prepare a smaller, CUDA-free backend:

cuda.ml::cuda_ml_install(device = "cpu")

The CPU backend can run a cuda.ml random forest trained on a GPU as well as a supported external tree ensemble. The complete installation can also run these models on a CPU, so the smaller backend is an optional deployment choice.

Save and deploy fitted models

cuda.ml 0.4.0 expands model persistence for training, analysis, and deployment workflows. Supported fitted models can be saved to a compressed file, restored in another R process, stored as raw bytes, or wrapped with the bundle package.

The simplest file workflow passes a path directly:

model <- cuda_ml_rand_forest(
  class ~ .,
  data = modeldata::hpc_data,
  trees = 100
)

cuda_ml_serialize(model, "hpc-runtime-forest.cuda-ml")

In another R process or deployment environment, prepare cuda.ml and restore the fitted model:

library(cuda.ml)

cuda_ml_install()
model <- cuda_ml_unserialize("hpc-runtime-forest.cuda-ml")

predict(model, modeldata::hpc_data[1:5, ], type = "class")

File paths use gzip compression. Passing connection = NULL instead returns the state as an uncompressed raw vector of bytes, which is convenient for object stores and database BLOB columns. The blob package can represent raw vectors for database workflows, and the bundle package is supported for teams that already use bundled model artifacts.

Persistence is available for linear models, logistic and multinomial regression, PCA, SVC and SVR models, UMAP, random forests, and nvForest models. cuda.ml checks the saved state and required backend before restoring it. For an nvForest-backed model, the restore call can select CPU or GPU inference.

Highlights for users upgrading from cuda.ml 0.3

This is a breaking update to the earlier package. The most visible changes are:

• The random-forest API now uses mtry for predictor sampling and sample_fraction for row sampling. trees defaults to 100, and an omitted seed draws from R’s random-number generator, so set.seed() controls the fit.
• Linear, logistic, and multinomial regression now use numeric penalty and mixture arguments that match parsnip. Logistic and multinomial regression are unregularized by default.
• normalize_input was removed from the linear-model functions. It previously requested GPU-side L2 normalization. Use explicit preprocessing such as recipes::step_normalize() when centering and scaling are appropriate; but please note, the two operations are not numerically identical.
• cuda_ml_sgd() now fits squared-loss regression only. Its loss argument was removed, and n_iters_no_change is now n_iter_no_change.
• The former FIL interface has been replaced by nvForest. Random projection and the KNN IVFSQ index have no replacement in the pinned upstream API.

See the full changelog for the complete list of API changes.

The guides cover first steps and more complete examples:

• Get started with cuda.ml
• Use cuda.ml with tidymodels
• Save and restore models
• nvForest inference and deployment

Python Polars: The Definitive Cheatsheet is out now, and it’s free. Download it, print it, laminate it, put it on your desk.

Get the cheatsheet

What’s on it

Page one covers the fundamentals: how to get started, data types, data structures, the eager and lazy APIs, reading and writing data, and transforming data with select, filter, sort, reshape, aggregate, and join.

Page two is mostly expressions, which is where Polars really shines. It also includes the most common methods for working with strings, datetimes, lists, structs, and categoricals, plus styling data with Great Tables and visualizing data with Altair and Plotnine.

Keep in mind that this is a lossy compression. Two pages can cover many functions and methods, but there’s no room for nuance or to explain the fundamental concepts. The cheatsheet is for remembering and referencing, not for learning Polars from scratch, and it doesn’t replace the official documentation or our book.

There’s an HTML version too

A laminated PDF is a good desk reference, but it isn’t much use to a screen reader, and you can’t copy anything out of it. So the cheatsheet also comes as HTML, with headings, tables, and selectable code. It works with a screen reader, the text reflows on a phone, and you can copy a snippet straight into your editor instead of retyping it.

Made together with Polars, Inc.

A cheatsheet is really a long series of arguments about what matters most. Every line that makes it displaces something else. Having Polars, Inc. involved meant the people who build the library had a say in what a two-page reference should cover, and the cheatsheet is considerably better for it.

Posit and Polars, Inc. are different companies, but our tools overlap. Great Tables and Plotnine are Posit projects, and both work directly with Polars DataFrames. That’s why they’re on the sheet, and it’s part of why this collaboration made sense.

We hope you find it useful. If there’s a method you think we wasted space on, or one we should have included, we’d like to hear about it.

Get the cheatsheet

Orbital’s whole pitch is that a trained model becomes SQL, so a database can run predictions on its own, with no Python process anywhere near it. Until 0.6.0, “trained model” meant scikit-learn: pipelines, trees, linear models, all .fit() in Python and then turned into a SELECT statement. It never covered what a lot of teams are actually training now: PyTorch models, not scikit-learn pipelines.

Orbital 0.6.0 closes that gap. A torch.nn.Sequential network, trained exactly the way you already train it, now compiles to the same kind of SQL a linear regression would.

No ONNX Runtime. No model server. No separate inference service to keep alive next to the database. Just a query.

Why does that work at all, for a framework Orbital was never written for? Because of a decision made long before PyTorch was ever on the table.

Why this isn’t a bolt-on

Orbital was never really a scikit-learn tool. Underneath, it converts a scikit-learn pipeline to ONNX (Open Neural Network Exchange, a standard graph format for trained models) using the skl2onnx library, then walks that graph node by node to produce SQL. Scikit-learn was always one hop removed from what Orbital actually translates.

Adding PyTorch meant taking the same hop from a different starting point. torch.onnx.export turns a torch.nn.Sequential model into that same kind of graph. Feed it into the translator that already existed, and the translator doesn’t know or care whether the graph came from PyTorch or scikit-learn.

The proof is in how little new code that took. The entire engine for running a feed-forward network in SQL is three small classes: a Gemm translator for Linear layers, ReLU, Sigmoid. Everything else (the translator base class, the optimizer, the per-dialect SQL compiler) already existed, built earlier for trees and linear models.

When I first thought of support for PyTorch, I put it this way: “the underlying translation works on ONNX graphs… the same value proposition orbital already provides for scikit-learn models can apply as it is to pytorch networks exported to ONNX.” That sentence turned out to be the whole implementation plan.

If you want the fuller picture of how a graph becomes SQL, parser, then translator, then optimizer, the architecture docs walk through all three stages in order.

That architecture is also why I keep calling this multiple frameworks, not two frameworks. Relu, Sigmoid, Tanh, and Softmax are single translators, not one per framework. Scikit-learn’s MLPClassifier and MLPRegressor reach them through MatMul and Add, PyTorch’s nn.Sequential reaches the exact same translators through Gemm. One implementation, two entry points, and no reason it has to stop at two.

Scikit-learn and PyTorch are both real and shipping today, which is enough on its own to call this “multiple frameworks.” But the dependency story is already moving that direction: issue #113 proposes turning scikit-learn itself into an optional dependency, the same way PyTorch already is, so the core stops assuming any particular framework at all.

Making it actually usable

Neural networks are layered, and every neuron in layer two reads every output of layer one. If Orbital just inlines those outputs at each place they’re read, instead of naming them once, that repetition compounds from one layer to the next. Two layers doubles the inlined text. Five layers is a different order of magnitude.

That’s not theoretical: the issue that tracked the fix measured scikit-learn’s own default MLPClassifier(hidden_layer_sizes=(100,)), the first thing anyone reaches for, at 53MB of generated SQL and roughly 894 seconds just to generate it. A hundred neurons in one hidden layer, and the query was already unusable.

PyTorch’s Gemm translator never had this problem. It has called preserve(), materializing its output as a real SQL column, since the day it was written. Scikit-learn’s MLPClassifier and MLPRegressor compile through different ONNX ops though: MatMul then Add, because that’s what skl2onnx emits, not Gemm. Neither of those translators called preserve() at all.

The fix, Optimizer.preserve_referenced_outputs(), runs after every single node in the translation loop, for every translator, not just MatMul and Add, and checks how many times that node’s output is actually referenced downstream. Referenced more than once, it gets materialized as a named column. Referenced once or not at all, it stays inlined, no extra column, no extra noise.

None of this is new machinery either. Tree ensembles already lean on the same trick: preserve() materializes per-tree votes, or the whole ensemble’s aggregated vote so it isn’t re-emitted everywhere it’s read, as real SQL columns. preserve_referenced_outputs() generalizes that same idea automatically, for every translator, whether it’s part of an ordinary pipeline or a neural network.

Here’s what that fix was worth, measured on three shapes while it was being built:

Generation time collapsed just as hard: the 20→64→64→1 network above went from about 185 seconds to about 3.6 seconds. Running the resulting SQL got faster too, if less dramatically: on 200,000 rows, an MLP(32,32) query dropped from 0.43s to 0.35s, and an MLP(100,100) from 3.58s to 3.31s.

Same hyperparameters. Same defaults everyone actually reaches for. The difference is entirely in how the SQL gets built, not in what the network computes.

What it can do today

Neural network support in 0.6.0 covers binary classification, multiclass classification, and regression, for both scikit-learn and PyTorch. Five new or updated examples in the repo prove it out: pytorch_fraud_detector.py, pytorch_maintenance_classifier.py, pytorch_demand_regressor.py, pipeline_mlp_classifier.py, pipeline_mlp_regressor.py.

The one worth walking through is the fraud detector, since it’s the shape most teams actually need: a handful of numeric features, a binary “is this fraud” output, trained the same way this kind of model always is.

FEATURES = {
    "amount": orbital.types.DoubleColumnType(),
    "hour": orbital.types.DoubleColumnType(),
    "v1": orbital.types.DoubleColumnType(),
    "v2": orbital.types.DoubleColumnType(),
}

model = torch.nn.Sequential(
    torch.nn.Linear(len(FEATURES), 16),
    torch.nn.ReLU(),
    torch.nn.Linear(16, 8),
    torch.nn.ReLU(),
    torch.nn.Linear(8, 1),
    torch.nn.Sigmoid(),
)

# ... train model normally: Adam, BCELoss, a plain training loop ...

orbital_pipeline = orbital.parse_pytorch_model(model, FEATURES)

From that one orbital_pipeline, two engines:

duckdb_sql = orbital.export_sql("transactions", orbital_pipeline, dialect="duckdb")
postgres_sql = orbital.export_sql("transactions", orbital_pipeline, dialect="postgres")

I ran both, against a real DuckDB and a real Postgres, not just read the generated text. Same four test transactions, three ways to compute a prediction (PyTorch itself, the DuckDB query, the Postgres query), and all three agree to within 1e-5. Both queries land around 10KB, not the tens of megabytes a naive translation would have produced before 0.6.0’s optimizer fix.

That works because Sigmoid and ReLU both compile to plain arithmetic: EXP, a division, a CASE WHEN. Every SQL engine has those. It’s not an accident which activation this example uses, either: Tanh compiles to a native TANH() call instead, and not every dialect implements that the same way, so it’s the one activation in Orbital’s NN support with an actual portability caveat attached.

DuckDB and Postgres are two of the three dialects Orbital actively tests in CI. SQLite is the third. But that’s a testing choice, not an architecture boundary: Orbital doesn’t write dialect-specific SQL at all. Translation ends at ibis. export_sql just hands the finished expression to whichever of ibis’s own backend compilers matches the dialect you ask for, and ibis ships about twenty of those: Snowflake, BigQuery, Trino, MySQL, and so on.

Limits, honestly

What Orbital 0.6.0 handles is feed-forward, fully connected networks: stacks of Linear layers with ReLU, Sigmoid, Tanh, or Softmax in between. That already covers real use cases people put into production: fraud scoring, churn, demand forecasting, risk models. None of those need a CNN or a transformer.

What it doesn’t do yet is exactly what that shape excludes: no convolutions, no recurrence, no attention, no embedding layers for categorical features. If your model needs any of those, Orbital isn’t there yet.

SQL size still grows with the network. A few hidden layers of 64 to 128 neurons land comfortably in the KB range. Wider or deeper than that, it’s worth checking the generated SQL against whatever statement-size limit your engine has, before you deploy it.

That headroom exists at all thanks to Optimizer.preserve_referenced_outputs(), and that mechanism helps every translator in Orbital, not just neural networks. It’s a big enough story on its own that I’ll get back to in a future post.

Try it

pip install orbital[pytorch]

From there, the getting-started guide walks through this same fraud-detector shape end to end, and the examples directory has five more, covering classification and regression in both frameworks.

Orbital speaks PyTorch now. Scikit-learn too. Same query either way.

There are more frameworks already in the works. I won’t name them here, only that the architecture was built for exactly this.

How do you know which model to use and when? Often, it’s not a question of which model is best, but of which model suits your task and needs at a given time. You might switch between models for different projects (package development vs. data analysis), or even within a single project (planning vs. implementation). Different tasks require a different mix of cost, token usage, speed, intelligence, and capabilities.

If you want our most durable, high-level advice: start with the most expensive model you have access to from either OpenAI or Anthropic.¹ Then, once you have a sense of the “ceiling,” try less expensive models and see how they compare. Developing a feel for what’s possible with LLMs will help you make better decisions about trade-offs.

That said, we’ll try to tackle this question more thoroughly in this post.

The model landscape

Currently, Anthropic and OpenAI set the bar for AI capabilities. Google is often discussed as a third frontier lab, though its current model lineup is less competitive.

Anthropic and OpenAI each release a “family” of models: Claude and GPT, respectively. The most capable models in both families are also the slowest and most expensive. Conversely, the least capable models are the cheapest and quickest. Other labs tend to follow this same pattern, releasing a set of models with different trade-offs along the cost-performance curve.

Models within a given family tend to share a similar shape of intelligence, with related capabilities (relative to model size), shortcomings, and idiosyncrasies. For example, Claude Fable 5, Claude Opus 5, and Claude Sonnet 5 often use the same turns of phrase and are quite good at writing code and debugging it. Models from different families can have different shapes of intelligence even when their benchmark scores and prices are very similar. For example, GPT-5.6 Terra and Claude Sonnet 5 are priced similarly and comparably capable at agentic coding, but Terra doesn’t “see” data visualizations as well as Sonnet, while Sonnet doesn’t communicate as clearly as Terra.

Other labs release models that score nearly as high as models from Anthropic and OpenAI on benchmarks. However, these evaluation scores can be deceiving. Labs can now train models to optimize for benchmarks (“benchmaxxing”). These models score well on benchmark-shaped tasks, which tend to be highly autonomous and “tricky,” but can fail to generalize to real-world tasks, which often involve more ambiguous requests. Kimi K3 and GLM 5.2 offer a counterexample: they score well on benchmarks, and we’ve also found them exceptionally well-rounded and intuitive. We recently introduced both models to Posit AI.

In previous newsletters and blog posts, we’ve shared results from targeted evaluations. Here, instead, we offer an approximate and entirely vibes-based comparison of these model families’ characteristics.

For data science applications broadly, you can loosely think of the relevant score as the average of the agentic coding and vision scores we’ve assigned here. Beyond writing R and Python code, models need to be able to interpret plots accurately and faithfully.

Find the model that fits your task

Much of model choice is constrained by the models you have access to. Your organization may only allow a certain provider, or you might not want to pay multiple (possibly expensive!) subscriptions just to have access to all the top models.

If you do have your pick, however, here are some quick guidelines, partially shaped by what models are currently available through Posit AI.

The best open-weights model, especially for data analysis: Kimi K3.

As Simon wrote in the Kimi K3 and GLM 5.2 in Posit AI announcement post, “Kimi K3 is currently the most capable open weights model out there. In our internal testing, it feels somewhere between Opus 5 and Fable 5, and is notably well-rounded compared to other open weights releases.”

It also ranks near the top in bluffbench2 (13.46%, compared with 16.35% for the tied top scorers, Gemini 3.5 Flash and Claude Fable 5), which evaluates models’ abilities to spot subtle data quality issues in visualizations.²

A highly autonomous model for a complex coding or data task when cost and speed aren’t a concern: Claude Opus 5, Claude Fable 5, or GPT-5.6 Sol.

These are the top-of-the-line models from Anthropic and OpenAI. They are expensive and relatively slow, but can be worth using for ambitious or highly autonomous work.

A strong open-weights model for coding when you don’t need vision: GLM 5.2 (from Z.ai).

“GLM 5.2 excels at agentic coding and less so at data analysis tasks.” It is much less expensive than the proprietary models it resembles for coding tasks, but it does not natively support image inputs.

A middle-tier generalist for coding or data analysis: Claude Sonnet 5 or GPT-5.6 Terra. Both are capable across coding and routine data analysis, support vision, and are less expensive than their respective labs’ higher-tier models.

Good plot or image interpretation: One of the Gemini 3.x Flash models (3.7 was released on August 13).

This model series is particularly strong at vision and has performed well on bluffbench (3.5 Flash) and bluffbench2 (3.5 and 3.6 Flash).

Fast answers from an Anthropic model for a task that is not particularly complex: Claude Haiku 4.5.

Fast answers from an open-weights model for a task that is not particularly complex: Gemma 4.

Assorted notes from August 2026

In late summer 2026, a few developments feel notable, but, as with much in the AI world, who knows how long these observations will hold.

• Google Gemini currently does not have any models that perform near the frontier. With the releases of Gemini 2.5 Pro (June 2025) and the Gemini 3 series (November 2025), Google seemed positioned as a third frontier lab. However, it’s been quite a while since they released a frontier model, and they’re now meaningfully behind. As of the time of writing, Google says Gemini 3.5 Pro is still testing with partners. Meanwhile, the company has begun training Gemini 4 and says it is excited by the progress.

• For a year or so, it seemed like Anthropic was solidly ahead of OpenAI in agentic coding. However, since the Claude 4.6 releases, it has become less clear that Anthropic is meaningfully ahead of OpenAI. For one, Anthropic’s current high-end models use newer tokenizers that, according to Anthropic, produce roughly 35% more tokens for the same text than their predecessors. This means the same listed price per token does not necessarily translate to the same cost for comparable text. Further, in our experience, the Claude series has become increasingly token-hungry and difficult to communicate with. At the same time, OpenAI’s models have a notably clear, concise communication style compared with the Claude 5 series. Anthropic is still likely ahead on autonomous, long-horizon coding, but OpenAI no longer feels behind for day-to-day software engineering and data science.³

• There are now a number of balanced, well-rounded open-weights models relatively close to the closed-weights frontier. Kimi K3 and GLM 5.2, in particular, combine strong capabilities with a more pleasant, intuitive feel than their predecessors. While earlier open-weights models were just as close to the frontier in benchmark scores, some newer releases are notably more well-rounded and respond to prompts about as effectively as proprietary models. These releases are also priced at a steep discount compared with the proprietary models they most resemble.

Recent past newsletters

• Keep track of your data exploration with Posit Assistant’s EDA log
• Which models are best at spotting subtle data quality problems?

Subscribe via RSS

Welcome back to another edition of our monthly Positron updates! Each month we share highlights from our latest release and useful resources. Last release we told you about several major features that came out of preview to general availability, including the new notebook editor. This milestone we are excited to share new functionality for SQL support, reproducible authoring, helping you know when packages are missing, and more.

Data Connections

Data Connections is our new way to work with SQL and database-like resources in Positron, from local files and database servers to cloud data warehouses. It is currently available as a preview feature, and you can enable it with the dataConnections.enabled setting. This release more than doubles the number of data sources you can reach. Amazon Redshift, Snowflake, Databricks, and Posit Connect pins join the existing DuckDB, PostgreSQL, and SQLite support.

The panel itself is more capable as well. Refresh and Refresh All reload the tree while preserving what you have expanded, briefly highlighting the rows that were reloaded. Open connections now show an indicator. Collapsing a connection in the UI keeps you connected to your data source, while anything you’ve previewed with the Data Explorer stays still open. When you remove a connection, Positron asks for confirmation and reports how many Data Explorers will close with it.

Data Connections is still an experimental preview, and your feedback continues to shape it. Tell us which databases and warehouses you need, and anything confusing, missing, or broken, in the Data Connections discussion.

Inline output for Quarto

Inline output for .qmd documents came out of preview last release, and this release brings you a substantial round of polish for this way of working. Be aware that the Quarto settings have moved into a dedicated quarto.* namespace with its own group in the Settings editor. The previous positron.quarto.* keys still work but are deprecated, and Positron will prompt you to update your settings.

Before a kernel starts, the kernel status names the interpreter it will start and offers an explicit Start Kernel action. When a cell fails, Fix and Explain buttons send the error to Posit Assistant, matching the Positron notebook experience. The editor also scrolls to reveal output as it is produced, which you can turn off with the new quarto.inlineOutput.autoScroll setting.

Output renders more faithfully as well. The editor gutter now shows which statement is currently executing and per-statement progress. Positron draws images at your display’s pixel ratio, so plots are sharp on retina screens, and Python figures now respect the fig-width and fig-height cell options.

HTML widgets no longer stick in the editor corner when you scroll past them, or trap scrolling instead of letting the document scroll. HTML widgets no longer render as raw HTML after a reload, and collapsed output no longer springs back open when its cell re-runs. Running code in a Quarto document also pins the editor tab now, so Positron does not silently close the document and its session when you open another file.

AI model providers

Positron now reads AI model provider configuration from a single providers.json file rather than a scattered set of settings. The new release will migrate your existing configuration automatically when you start it, and deprecates the authentication.* and positron.assistant.provider.*.enable settings in favor of it. Two new commands give you direct access: Open AI Provider Settings (JSON) opens providers.json from the Command Palette, and Migrate Provider Settings to providers.json runs the migration on demand.

Install missing packages

Positron now notices when your code refers to a package you do not have installed and offers to install it for you. The prompt appears for packages referenced by your scripts and notebooks in both R and Python.

A library() or import call for something missing becomes a single click instead of an error you have to go resolve by hand.

Performance and memory

We continue to invest in the memory footprint, performance, and reliability of Positron. Several components now load only when they are actually needed, and turning off ai.enabled now means Positron never loads some heavy AI-related components at all. We fixed a cluster of long-standing reliability problems around session restarts and lifecycles.

Startup and editing are faster as well. R and Python kernels start much faster on Windows systems with aggressive antivirus software. The interpreter session picker appears immediately instead of waiting for interpreter discovery to finish. We also fixed slow typing, formatting, and saving in long R and Python files.

What’s coming next

• Join our upcoming webinar on August 26 to learn about AI governance in Posit Workbench.
• We are looking forward to posit::conf(2026) next month, where our team will have several sessions on Positron. Register now to join us in person in Houston or virtually from anywhere in the world.

install.packages("themis")

This release provides a substantial amount of new features. We will cover the highlights in this blog post, which include: 11 new steps, setting sampling targets per class, and more distance metrics. See the news file for a complete list of changes in this release.

To get started, load the tidymodels and themis packages:

library(tidymodels)
library(themis)

set.seed(1234)

New sampling steps

This release adds support for 11 new steps for under- and over-sampling of data. Making for a more complete picture of the commonly discussed methods in this category. Each step also ships a direct-implementation counterpart (enn(), smogn(), and so on) for use outside a recipe.

The new steps fall into four groups.

Cleaning-based under-sampling removes observations that sit in the wrong neighborhood:

• step_enn() applies the Edited Nearest Neighbors rule.
• step_ncl() applies the Neighborhood Cleaning Rule.
• step_cnn() uses Condensed Nearest Neighbors.
• step_oss() uses One-Sided Selection.

Selection-based under-sampling picks a smaller set of representatives instead:

• step_cluster_centroids() replaces each class with one representative per k-means cluster.
• step_instance_hardness() removes the observations that are hardest to classify.

Over-sampling gains three new variants:

• step_kmeans_smote() generates new examples only inside clusters where the minority class dominates.
• step_svmsmote() concentrates them near the decision boundary found by a support vector machine.
• step_smoten() handles data where every predictor is categorical.

Lastly we have a step that applies the SMOTE idea to regression:

• step_smogn() over-samples rare regions of a numeric outcome and under-samples the common ones.

The new “Methods overview” article lays out the full taxonomy.

Sampling targets per class

The over_ratio and under_ratio arguments used by many of the steps in themis previously only took a single value. We now accept a named vector, such that you can specify the value for each level of the outcome.

The penguins data set has three species of unequal size. A few penguins have missing measurements, which we drop up front so the counts below are easier to follow:

data(penguins, package = "modeldata")

penguins <- penguins |>
  drop_na()

count(penguins, species)

# A tibble: 3 × 2
  species       n
  <fct>     <int>
1 Adelie      146
2 Chinstrap    68
3 Gentoo      119

The names of the vector should correspond to the levels of the outcome. So over_ratio = c(Chinstrap = 0.8, Gentoo = 1) brings Gentoo up to the size of the majority level, and Chinstrap up to 80% of it:

recipe(species ~ bill_length_mm + bill_depth_mm + flipper_length_mm + body_mass_g,
       data = penguins) |>
  step_smote(species, over_ratio = c(Chinstrap = 0.8, Gentoo = 1)) |>
  prep() |>
  bake(new_data = NULL) |>
  count(species)

# A tibble: 3 × 2
  species       n
  <fct>     <int>
1 Adelie      146
2 Chinstrap   117
3 Gentoo      146

Levels you don’t name are left untouched, as are rows with a missing outcome. Two things to keep in mind: supplying a vector means the argument can no longer be tuned, and step_rose() still requires a single number, because its over_ratio scales the size of the total generated sample rather than setting a per-class target. The new article on over_ratio and under_ratio walks through both arguments in more detail.

Tracking synthetic rows

Every up-sampling step gains an indicator_column argument. Give it a name and the final data gets a logical column marking which rows the step added:

smote_res <- recipe(class ~ x + y, data = circle_example) |>
  step_smote(class, indicator_column = "synthetic") |>
  prep() |>
  bake(new_data = NULL)

count(smote_res, class, synthetic)

# A tibble: 3 × 3
  class  synthetic     n
  <fct>  <lgl>     <int>
1 Circle FALSE        58
2 Circle TRUE        284
3 Rest   FALSE       342

This will mostly be useful as a diagnostic tool, or to help visualize how these methods work in practice.

smote_res |>
  ggplot(aes(x, y, color = synthetic)) +
  geom_point(alpha = 0.7) +
  labs(
    title = "Synthetic minority observations created by step_smote()",
    color = "Synthetic"
  ) +
  theme_minimal()

More distance metrics

Most of the steps in this package are built on some calculation that has to do with nearest neighbors. And so far all of them had been using Euclidean distances. We have added a distance argument to every step that deals with neighbors, letting you choose a different distance metric.

manhattan_res <- recipe(class ~ x + y, data = circle_example) |>
  step_smote(class, distance = "manhattan", indicator_column = "synthetic") |>
  prep() |>
  bake(new_data = NULL)

count(manhattan_res, class, synthetic)

# A tibble: 3 × 3
  class  synthetic     n
  <fct>  <lgl>     <int>
1 Circle FALSE        58
2 Circle TRUE        284
3 Rest   FALSE       342

The class counts are unchanged, since those are set by over_ratio rather than by the metric. What changes is where the synthetic observations land, because a different metric picks different nearest neighbors to interpolate between.

Acknowledgements

A big thank you to everyone who has contributed issues, pull requests, and discussion since the last release! @3styleJam, @Dodothereal, @EmilHvitfeldt, @FvD, @jeroenjanssens, @SAY-5, and @topepo.