satelle

AI SDLC for developers

Your agent writes the code.
 satelle runs your process.

satelle is an AI SDLC for developers. You write your process — stories, steps, gates — as files in your repo. Your CLI agent follows it, and a separate reviewer decides when a story is done. Runs locally, one binary. 

curl -fsSL https://github.com/bobmcallan/satelle/releases/latest/download/install.sh | sh

runs locally · one static binary · your terminal agent · your subscription

Install satelle How it works GitHub →
claude code — driving a satelle story
A CLI agent drives a story through the workflow in this repo. Each gate is judged by a separate reviewer.

How a story moves

This is the default satelle init gives you, taken from the satelle repo. It is the minimum, not the process.

  1. backlog A story is raised: a title, a goal, numbered acceptance criteria.
  2. in_progress Entry is gated by an intent review: is the story well-formed enough to work on? Your agent does the work.
  3. done Entry is gated by a scope review (did the change stay inside the story?) and a done review (are the acceptance criteria met?). Status moves only through an accept.

No plan step, no release step, no opinion on how the work is done — that is yours to author.

Build the process you need

A process is a route: steps, who performs each, and what gates entry. Add steps, drop steps, gate them with a reviewer rubric or a command, run each step on a different agent. This is the route this site is built through. 

stage         performed by              gated on entry by
backlog       raised
plan          planner writes the plan
in_progress   executor codes            plan review
integration   executor integrates       acceptance-criteria review · design review when the story is tagged UI
deploy        executor deploys locally  full test suite (a command; exit 0 accepts)
release       executor pushes           deploy check (a command)
done                                    release review

Each of those is a file under .satelle/ in this repo: the route, the skills each step and gate run, and the principles the agent reads first.

Your agent, your subscription

satelle does not ship an agent or sell tokens. It drives the CLI agent you already use — claude and grok are the ones we run — over its command line or ACP. Each step can name a different agent, model and reasoning effort, so judgment runs where it pays and execution runs cheap.

What it leaves alone

  1. Your machine One static binary over SQLite in your repo; it never phones home.
  2. The agent loop satelle does not run the agent; it drives the one you configure.
  3. Its own grading The executor cannot pass a gate it is standing in.

Get started

Install satelle Read the docs