A team, not an assistant
A chief of staff you talk to, a Planning Lead that scopes the work, a Builder that writes it, and a Reviewer that reads it back. Roles with mandates and tool policies, rather than one agent with a longer prompt.
Agent teams · on your own machine
Everything your runtime can do, plus a team and a brain.
Lucitra runs a team of AI agents on your own computer, on the runtime subscription you already pay for. They share a memory that lives in your own Git repository — one you can read, diff, and approve changes to — and they finish by opening a pull request that you merge.
Not released yet. Leave an address and we will tell you when there is something to install — no account, and nothing else sent here.
A different model per role
Each agent runs on the model you pick for it. Your Reviewer does not have to be the model that wrote the code — which is the one review property no single-vendor assistant can offer.
A brain you can read
Project memory kept as Markdown in a Git repository you own. You read it, you diff it, and an agent changes it the way anything else changes: by opening a pull request you approve.
Your runtime, your machine
Agents run on your computer, on the Claude Code, Codex or goose subscription you already pay for. We do not resell model capacity and we do not meter your usage.
Done means a pull request
Work happens in an isolated copy of your project and finishes by opening a PR. Nothing merges without a person — not in any mode, not by any agent.