Jarvis is an autonomous software engineer that connects to your Linear board and GitHub repos. It researches, plans, implements, reviews, and merges code — with human approval at every checkpoint you choose.
Jarvis watches your Linear board and autonomously processes tickets through a six-stage pipeline.
Explores the codebase, analyzes related files, and posts structured findings to the ticket.
Reads the research and designs a phased implementation plan with file changes and test strategy.
You review the research and plan — approve to proceed or reject for revision.
Creates a branch, writes code (TDD), runs tests, captures visual proof, and pushes the PR.
Validates against acceptance criteria, runs tests, checks scope, and delivers PASS/FAIL verdict.
Final human check — approve the merge or send back for changes.
Writes retrospective lessons to the knowledge base and merges the PR.
gates_enabled = false for full autonomy.
Jarvis isn't a demo — it's a production system used to ship real code.
Bi-directional sync with your Linear board. Jarvis reads tickets, updates states, posts comments, and attaches artifacts at every stage.
Each ticket gets its own branch and worktree. Code is pushed, reviewed, and merged automatically — fully traceable.
Runs on Claude Code with three configurable model tiers: fast (haiku), normal (sonnet), and advanced (opus). Swap providers without touching code.
Approve or reject at every checkpoint via Slack or Telegram. Go fully autonomous with a single config toggle — gates_enabled = false.
Define your own pipeline stages, gates, and prompts in a YAML file. Add, remove, or reorder stages without touching code.
Monitor active agents, recent outcomes, runtime, and idle clocks. SSE-powered live updates — pause, cancel, or retry with one click.
Describe an idea in Slack — the bot refines it into a Linear ticket through a short chat. Approve with a button and Jarvis takes it from there.
Ephemeral worker containers on EC2 spot instances. Pay only for active work — infrastructure scales down to nothing when idle.
Run Jarvis across multiple repos and Linear teams simultaneously. Per-project config for language, test commands, and branch conventions.
Jarvis is in active development and accepting waitlist signups. Early adopters get priority access, dedicated support, and a say in the roadmap.
Share Jarvis with your network and earn 30% of subscription revenue from every developer you bring in.
You earn 30% of every subscription payment from people who sign up through your referral link — for as long as they're customers.
Refer as many people as you want. There's no upper limit on your earnings. Each referral is tracked to your unique code forever.
Earnings are calculated and paid out every quarter. You get a full breakdown of your referrals and their status.
Every signup gets a personal referral link. Your code is embedded in the URL — no accounts to manage, just share and earn.
Yes — Jarvis v0.1 is running in production. It's open source and you can self-host it. The waitlist is for early access to hosted plans and priority support.
Jarvis runs on Claude Code with three configurable model tiers. You can also swap providers via OpenRouter or bring your own Anthropic/ZAI API keys. The [models] section in jarvis.toml makes it a one-line change.
Absolutely. Drop a flow.yaml in your config directory and define custom stages, gates, and prompts. Add, remove, or reorder stages — the full pipeline is YAML-configurable. See examples/flow.yaml in the repo.
When someone signs up using your referral link, you earn 30% of their subscription fees for as long as they stay on a paid plan. There's no cap — refer 10 developers and earn from all 10. Payouts happen quarterly.
Yes. The [[projects]] config supports multiple repos, each with its own Linear team, branch conventions, language profile, and test commands. They all run in the same daemon process.
Jarvis runs as a single binary with a Claude Code subscription (for the agent) and Docker (for the runner). EC2 spot instances are optional for scale-to-zero — you can run it on a single VM or your laptop.