RStudio-native AI copilot for serious R analysis.
It sees your code, models, objects, git changes, and reporting workflow, then helps like an analyst instead of a generic chatbot.

Install from CRAN Package website Reference Issues

Ravel is not just chat inside an IDE. It is designed to behave like an analysis copilot for R users: it understands the active script, selected code, loaded objects, model outputs, git changes, and reproducible reporting workflows so it can help with real RStudio work instead of acting like a generic web chatbot.

Stable release from CRAN

install.packages("ravel")
library(ravel)
ravel::ravel_setup_addin()

Development version from GitHub

if (!requireNamespace("pak", quietly = TRUE)) install.packages("pak")
pak::pak("msaule/ravel")
library(ravel)
ravel::ravel_setup_addin()

1. Install ravel from CRAN.
2. Run ravel::ravel_setup_addin() to connect at least one provider.
3. Run ravel::ravel_chat_addin() in RStudio to open the chat UI.

• Context-aware by default. Ravel gathers the active editor, selected code, workspace objects, console state captured through Ravel actions, project files, and recent git diffs before it responds.
• Built for statistical work. It explains lm() and glm() results, coefficients, interactions, fit diagnostics, and common modeling pitfalls in plain English.
• Safe when it acts. Generated code is previewed, file edits are staged, and actions are logged instead of silently executed.
• Designed for RStudio. The setup flow, chat UI, and action workflow live inside RStudio addins rather than treating R as a thin wrapper around a generic chat window.
• Multi-provider without pretending. Ravel supports official APIs and official CLIs only, with clear messaging when a provider or auth path is unavailable.

These are the kinds of prompts where Ravel starts feeling very different from a normal chat window.

Why do these logistic coefficients explode?
Check whether separation is plausible,
tell me what diagnostics to run next,
and draft a Quarto diagnostics subsection.

Convert this dplyr pipeline to base R,
keep the same output shape,
and explain each transformation step.

Write a results section from this model,
explain the interaction in plain English,
and include a Quarto chunk for follow-up plots.

Summarize what changed in this analysis,
tell me which edits could change results,
and suggest a validation checklist before merge.

Explain these coefficients like I have to present them.
What is the reference group?
How does the interaction change the main-effect interpretation?

Why is this join failing right now?
Use the selected code and loaded objects,
then propose the smallest safe fix.

• Active editor contents and selected code
• Loaded objects, including data frames, formulas, and fitted models
• Ravel-managed console output and current session details
• Project files, working directory, and package context
• Workspace and editor git state, including recent diffs

• Explain selected R code and debugging errors
• Interpret model summaries, coefficients, interactions, and diagnostics
• Compare modeling choices and suggest next checks
• Refactor tidyverse and base R code in either direction
• Draft Quarto methods, results, and diagnostics sections
• Preview code and file actions before applying them

• Analysts working in RStudio on scripts, reports, and iterative modeling
• Students learning how code, model output, and interpretation connect
• Researchers writing Quarto or R Markdown from live analysis objects
• Data science teams reviewing analytical changes, not just syntax
• Anyone who wants safer execution than copy-pasting from a browser chatbot

Ravel is explicit about what is supported today and what is still constrained by official provider boundaries.

• No silent code execution by default
• No silent file edits by default
• Explicit previews and approvals
• Structured history for actions and conversations, stored in session memory by default
• Honest provider and auth messaging
• Statistical caveats when assumptions or limitations are visible

Non-sensitive settings and history stay in session memory by default, so Ravel does not write into a user's home filespace unless storage paths are configured explicitly through options(ravel.user_dirs = list(config = "<path>", data = "<path>")).

• CRAN package page: https://cran.r-project.org/package=ravel
• Package website: https://msaule.github.io/ravel/
• Showcase article: https://msaule.github.io/ravel/articles/ravel-showcase.html
• ARCHITECTURE.md explains the layers and execution model.
• ROADMAP.md lays out the planned phases beyond the MVP.
• CONTRIBUTING.md explains the developer workflow and release checks.
• RELEASING.md captures the CRAN and R-universe release path.
• CODE_OF_CONDUCT.md describes community expectations.
• AGENTS.md describes collaboration conventions for contributors and coding agents.

If you are developing on the repository locally, prefer:

devtools::load_all(".")

Use devtools::install(".") only when you specifically need the installed package. Full release and submission details live in RELEASING.md.

The auth and provider boundaries in this project follow official documentation:

• OpenAI Codex CLI: https://developers.openai.com/codex/cli
• OpenAI API auth: https://developers.openai.com/api/reference/overview
• GitHub CLI gh copilot: https://cli.github.com/manual/gh_copilot
• Gemini API docs: https://ai.google.dev/gemini-api/docs
• Anthropic API docs: https://docs.anthropic.com/en/api