Codewhale
A terminal coding agent, written in Rust, that edits files and runs commands with any AI model.
- Stars
- 41.2k
- Forks
- 3.6k
- Licence
- MIT
- Language
- Rust
- Last update
- 4 days ago
- Type
- CLI
What it is
Codewhale is an open-source coding agent you run in your terminal. You give it a task, and it reads your git repository, edits files, and runs commands until the task is done or it needs your input. It works with more than 40 built-in providers and with local models, and its /receipts command lists the files changed, commands run, and web and MCP calls made.
- Ask it to fix failing tests and explain what changed
- Run one task non-interactively with codewhale exec in a script
- Review a list of changed files and commands with /receipts
- Undo the last request and its file changes with /undo
- Switch models between turns with /model
Great for
- Developers who want an AI agent working inside a terminal and a git repository
- People who want to switch between model providers or use a local model
- Users who want a record of what the agent changed and ran
- Teams that want to run one task non-interactively in scripts or CI via codewhale exec
Not ideal for
- People who want a point-and-click app with no terminal, since it is a command-line tool
- Anyone who cannot review diffs or approve shell commands, because the default settings apply file edits without asking and Full Access turns off the sandbox
- Users who need guaranteed results, since the README does not describe accuracy or benchmarks
How to get it
- 1
See the README
You'll need: Git (the agent is meant to run from a git repository) · An API key for a chosen provider, or a local runtime such as Ollama
Install on macOS and Linux
curl -fsSL https://codewhale.net/install.sh | shInstall via npm
npm install -g codewhaleInstall via Cargo
cargo install codewhale-cli --locked - 2
Run it
- Commit or stash your work, then run codewhale from the root of a git repository
- Run /provider (or press F3), pick a provider, paste its API key, and choose a model
- Type a task such as: Fix the failing tests and explain what changed
- Approve or refuse shell commands when prompted (y, a, or n)
- 3
Get the most out of it
- Use /receipts or codewhale receipts --last to review what the agent did
- Use /undo to take back your last request and its file changes
- Use /model auto to pick a model per turn (set up with /router)
- Use Plan mode to have the agent read and search without changing files
- Use --sandbox workspace-write with codewhale exec to keep the shell sandbox on
Downloads
Strengths
- Supports many providers, including OpenAI-compatible endpoints and local runtimes like Ollama
- Records files changed, commands run, and web and MCP calls through /receipts
- Asks before running non-read-only shell commands in the default Ask permission
Trade-offs
- Default settings apply file edits without asking, and Full Access disables the shell sandbox, so careful permission choices matter
- Some features are marked experimental or tied to specific releases, such as the ChatGPT plan login
- The README is truncated here, so setup details beyond the install command are limited in this summary
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| Codewhale (this) | 41.2k | CLI | intermediate | 4 days ago | Developers who want an AI agent working inside a terminal and a git repository |
| opencode | 224k | AI model / agent | intermediate | 4 days ago | Developers who want an AI agent that works inside their terminal and existing projects |
| jcode | 21.4k | CLI | intermediate | 4 days ago | Developers who want several AI coding agents working on one repo at the same time |
| paperclip | 110k | App | advanced | 4 days ago | Developers already running several AI coding agents who want one place to track them |
| Vibe-Trading | 37.4k | AI model / agent | intermediate | 4 days ago | Developers who want an AI agent for quantitative trading research |
| invisible_dots | 32.4k | AI model / agent | intermediate | 6 days ago | Developers who want a self-hosted agent and are comfortable with a command-line setup |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.