01
Point it at an issue
Write the issue normally. Add one exact comment:
@louibot implement. If a product decision is missing,
Louibot pauses and asks focused questions first.
Local-first issue implementation
Assign the issue.
Get a draft PR.
Louibot is a small, transparent worker for the issues your team already writes. It runs on your machine, works in an isolated sandbox, and leaves the final decision with a human.
louibot / local
$ louibot doctor
✓ GitHub access
✓ Docker sandbox
✓ Model provider
✓ Repository checks
$ louibot start
issue #42 waiting for @louibot implement
→ worktree ready / agent listening
01Runs locally
02Draft PRs only
03Human review stays in charge
The loop
A narrow path from issue to review.
Louibot handles the repetitive middle while GitHub remains the collaboration surface your team knows.
02
It prepares the workspace
A persistent Git worktree and disposable Docker sandbox keep the host and credentials out of the coding loop.
03
The agent does the work
It reads repository instructions, edits through one shell tool, installs dependencies, and runs discovered checks.
04
You review the draft
Louibot preserves your repository PR template, adds its summary
and validation, and opens a draft. Submit inline feedback, then
comment @louibot revise to continue the same work.
Why this shape
Useful without pretending to be autonomous.
Louibot is opinionated about the boundary: it can implement a small, well-defined issue, but it cannot invent product requirements, merge code, or quietly operate a server on your behalf.
↳
Local execution. Your repository, API keys, checkpoints, and logs stay on your machine.
↳
Explicit triggers. Humans start implementation and revisions with visible GitHub comments.
↳
Inspectable state. A small CLI shows active jobs, logs, costs, pauses, and failures.
Get started
Three commands to your first worker.
Bring Node.js 22+, GitHub CLI, Docker, and an API key for a tool-calling model. The rest lives in the repository.
terminalcopy / paste
# install
npm install --global louibot
# configure from your repository root
louibot init --model openai/YOUR_MODEL
louibot doctor
# start polling GitHub
louibot start
A
Use your own model key
OpenAI, Anthropic, and OpenAI-compatible Chat Completions endpoints are supported. Keys remain in the host environment.
B
Keep the issue small
Good inputs have a clear outcome, nearby tests, and repository
instructions in AGENTS.md.
C
Stay in control
Louibot creates draft PRs and never merges them. Stop, inspect, retry, or close the work from the CLI or GitHub.
See the implementation
Everything is in the open.
The worker, workflow, sandbox boundary, and tests are public. The todo repository shows a clarification round and a review-driven revision in real GitHub issues and draft PRs.