https://github.com/mlverse/chattr
https://github.com/mlverse/chattr
Last synced: 6 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/mlverse/chattr
- Owner: mlverse
- License: other
- Created: 2023-03-22T14:58:30.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2025-05-05T19:20:52.000Z (6 months ago)
- Last Synced: 2025-05-14T07:09:18.644Z (6 months ago)
- Language: R
- Homepage: https://mlverse.github.io/chattr/
- Size: 12.2 MB
- Stars: 229
- Watchers: 7
- Forks: 23
- Open Issues: 33
-
Metadata Files:
- Readme: README.md
- Changelog: NEWS.md
- License: LICENSE
Awesome Lists containing this project
- awesome-shiny-extensions - chattr - Interact with large language models (LLMs) in RStudio through the R console or Shiny gadget. (UI Components / Chat)
- jimsghstars - mlverse/chattr - (R)
README
# chattr
[](https://github.com/mlverse/chattr/actions/workflows/R-CMD-check.yaml)
[](https://app.codecov.io/gh/mlverse/chattr?branch=main)
[](https://CRAN.R-project.org/package=chattr)
[](https://lifecycle.r-lib.org/articles/stages.html#experimental)
- [Intro](#intro)
- [Install](#install)
- [Using](#using)
- [Available models](#available-models)
- [The App](#the-app)
- [Additional ways to interact](#additional-ways-to-interact)
- [How it works](#how-it-works)
- [Keyboard Shortcut](#keyboard-shortcut)
- [How to setup the keyboard
shortcut](#how-to-setup-the-keyboard-shortcut)
## Intro
`chattr` is an interface to LLMs (Large Language Models). It enables
interaction with the model directly from RStudio and Positron. `chattr`
allows you to submit a prompt to the LLM from your script, or by using
the provided Shiny Gadget.
This package’s main goal is to aid in exploratory data analysis (EDA)
tasks. The additional information appended to your request, provides a
sort of “guard rails”, so that the packages and techniques we usually
recommend as best practice, are used in the model’s responses.
## Install
To install the CRAN version of this package use:
``` r
install.packages("chattr")
```
If you wish to use the development version use:
``` r
pak::pak("mlverse/chattr")
```
## Using
Starting with version 0.3, `chattr` integrates with LLM’s via the
[`ellmer`](https://ellmer.tidyverse.org/) package. `ellmer` has a
growing list of LLM integrations, including
[OpenAI](https://ellmer.tidyverse.org/reference/chat_openai.html),
[Gemini](https://ellmer.tidyverse.org/reference/chat_gemini.html),
[Deepseek](https://ellmer.tidyverse.org/reference/chat_deepseek.html)
and others.
There are several ways to let `chattr` know which LLM to use:
- **Pre-set an R option** - Pass the `ellmer` connection command you
wish to use in the `.chattr_chat` option, for example:
`options(.chattr_chat = ellmer::chat_claude())`. If you add that
code to your *.Rprofile*, `chattr` will use that as the default
model and settings to use every time you start an R session. Use the
`usethis::edit_r_profile()` command to easily edit your *.Rprofile*
- **Use an `ellmer` object** - You can pass an `ellmer` chat object
directly to `chattr_use()`:
``` r
my_chat <- ellmer::chat_claude()
chattr_use(my_chat)
```
- **Named model** - You pass one of several pre-defined provider/model
setups. These setups are represented by labels set by `chattr`. At
this time, the combinations cover 3 providers: OpenAI, Databricks,
and Ollama. To use, simply pass the label to `chattr_use`. To see a
full list of the available pre-defined combinations and their
argument values see [Available models](#available-models). For
example, if you wish to use OpenAI’s GPT 4.1 Nano model, you simply
pass the corresponding label:
``` r
chattr_use("gpt41-nano")
```
- **Select one from a menu (legacy)** - If nothing is passed to
`chattr_use()`, and no option is set, `chattr` will try to create
the `ellmer` chat for you. It will try to figure if you have
authentication tokens for **OpenAI**, **Databricks**, and checks if
**Ollama** is running on your machine. `chattr` then returns a menu
based on the providers it was able to find:
``` r
chattr_use()
── chattr - Available models
Select the number of the model you would like to use:
1: Databricks - databricks-dbrx-instruct (databricks-dbrx)
2: Databricks - databricks-meta-llama-3-3-70b-instruct (databricks-meta-llama31-70b)
3: Databricks - databricks-mixtral-8x7b-instruct (databricks-mixtral8x7b)
4: OpenAI - Chat Completions - gpt-4.1-mini (gpt41-mini)
5: OpenAI - Chat Completions - gpt-4.1-nano (gpt41-nano)
6: OpenAI - Chat Completions - gpt-4.1 (gpt41)
7: OpenAI - Chat Completions - gpt-4o (gpt4o)
8: Ollama - llama3.2 (ollama)
Selection:
```
### Available models
For convenience, `chattr` contains some provider/model combinations that
you can use by passing what is under **Use value** to `chattr_use()`:
Model & Provider
Use value
DBRX (Databricks)
databricks-dbrx
Meta Llama 3.3 70B (Databricks)
databricks-meta-llama31-70b
Mixtral 8x7b (Datbricks)
databricks-mixtral8x7b
GPT 4.1 Mini (OpenAI)
gpt41-mini
GPT 4.1 Nano (OpenAI)
gpt41-nano
GPT 4.1 (OpenAI)
gpt41
GPT 4 Omni (OpenAI)
gpt4o
Llama 3.2 (Ollama)
ollama
If the provider and/or model you wish to use is not listed in the table
above, you can create an `ellmer` chat connection directly. And then
pass that chat object to `chattr_use()`. Here is a list of the providers
that are currently available in that package:
- Anthropic’s Claude:
[`ellmer::chat_claude()`](https://ellmer.tidyverse.org/reference/chat_claude.html)
- AWS Bedrock:
[`ellmer::chat_bedrock()`](https://ellmer.tidyverse.org/reference/chat_bedrock.html)
- Azure OpenAI:
[`ellmer::chat_azure()`](https://ellmer.tidyverse.org/reference/chat_azure.html)
- Databricks:
[`ellmer::chat_databricks()`](https://ellmer.tidyverse.org/reference/chat_databricks.html)
- DeepSeek:
[`ellmer::chat_deepseek()`](https://ellmer.tidyverse.org/reference/chat_deepseek.html)
- GitHub model marketplace:
[`ellmer::chat_github()`](https://ellmer.tidyverse.org/reference/chat_github.html)
- Google Gemini:
[`ellmer::chat_gemini()`](https://ellmer.tidyverse.org/reference/chat_gemini.html)
- Groq:
[`ellmer::chat_groq()`](https://ellmer.tidyverse.org/reference/chat_groq.html)
- Ollama:
[`ellmer::chat_ollama()`](https://ellmer.tidyverse.org/reference/chat_ollama.html)
- OpenAI:
[`ellmer::chat_openai()`](https://ellmer.tidyverse.org/reference/chat_openai.html)
- OpenRouter:
[`ellmer::chat_openrouter()`](https://ellmer.tidyverse.org/reference/chat_openrouter.html)
- perplexity.ai:
[`ellmer::chat_perplexity()`](https://ellmer.tidyverse.org/reference/chat_perplexity.html)
- Snowflake Cortex:
[`ellmer::chat_snowflake()`](https://ellmer.tidyverse.org/reference/chat_snowflake.html)
- VLLM:
[`ellmer::chat_cortex_analyst()`](https://ellmer.tidyverse.org/reference/chat_cortex_analyst.html)
### The App
The main way to use `chattr` is through the Shiny Gadget app. By
default, in RStudio the app will run inside the Viewer pane.
``` r
chattr_use("ollama")
chattr_app()
```
Screenshot of the Sniny gadget app in a
dark mode RStudio theme
After the LLM finishes its response, the `chattr` app processes all
markdown code chunks. It will place three convenience buttons:
- **Copy to clipboard** - It will write the code inside the chunk to
your clipboard.
- **Copy to document** - It will copy-paste the code directly to where
the app was called from. If the app is started while working on a
script, `chattr` will copy the code to that same script.
- **Copy to new script** - It creates a new R script in the RStudio
IDE, and copies the content of the chunk directly to it. Very useful
when the LLM writes a Shiny app for you
A lot of effort was put in to make the app’s appearance as close as
possible to the IDE. This way it feels more integrated with your work
space. This includes switching the color scheme based on the current
RStudio theme being light, or dark.
The settings screen can be accessed by clicking on the “gear” button.
The screen that opens will contain the following:
- Save and Open chats - This is an early experiment to allow us to
save and retrieve past chats. `chattr` will save the file in an RDS
format. The main objective of this feature, is to be able to see
past chats, not to continue previous conversations with the LLM.
- Prompt settings - In this section you can change the additional
information attached to your prompt. Including the number of max
data files, and data frames sent to the LLM.
Screenshot of the Sniny gadget
options
### Additional ways to interact
Apart from the Shiny app, `chattr` provides two more ways to interact
with the LLM. For details, see: [Other
interfaces](https://mlverse.github.io/chattr/articles/other-interfaces.html)
## How it works
`chattr` enriches your request with additional instructions, name and
structure of data frames currently in your environment, the path for the
data files in your working directory. If supported by the model,
`chattr` will include the current chat history.
Diagram that illustrates how
chattr handles model requests
To see what `chattr` will send to the model, set the `preview` argument
to `TRUE`:
``` r
library(chattr)
data(mtcars)
data(iris)
chattr_use("gpt4o")
#>
#> ── chattr
#> • Provider: OpenAI - Chat Completions
#> • Model: gpt-4o
#> • Label: GPT 4 Omni (OpenAI)
chattr(preview = TRUE)
#>
#> ── chattr ──────────────────────────────────────────────────────────────────────
#>
#> ── Preview for: Console
#> • Provider: OpenAI - Chat Completions
#> • Model: gpt-4o
#> • Label: GPT 4 Omni (OpenAI)
#> • temperature: 0.01
#> • max_tokens: 1000
#> • stream: TRUE
#>
#> ── Prompt:
#> [Your future prompt goes here]
```
## Keyboard Shortcut
The best way to access `chattr`’s app is by setting up a keyboard
shortcut for it. This package includes an RStudio Addin that gives us
direct access to the app, which in turn, allows a **keyboard shortcut**
to be assigned to the addin. The name of the addin is: “Open Chat”. If
you are not familiar with how to assign a keyboard shortcut see the next
section.
### How to setup the keyboard shortcut
- Select *Tools* in the top menu, and then select *Modify Keyboard
Shortcuts*
Screenshot that shows where to find the
option to modify the keyboard shortcuts
- Search for the `chattr` adding by writing “open chat”, in the search
box
Screenshot that shows where to input the
addin search
- To select a key combination for your shortcut, click on the Shortcut
box and then type *press* the key combination in your keyboard. In
my case, I chose *Ctrl+Shift+C*
Screenshot that shows what the interface
looks like when a shortcut has been selected