Syrus is an open-source, self-hosted automation harness for GitHub-based AI coding workflows. It turns issues, chats, and narrated video walkthroughs into tracked pull requests while keeping repositories, credentials, checks, transcripts, review policy, and merge flow in infrastructure you control.
Bis dat qui cito dat. “He gives twice who gives quickly.” — Publilius Syrus
How work flows through Syrus
Every goal turns into tracked epics and jobs, so you can see what's in progress, what's in review, and what shipped — and the merge still waits for the review policy you set.
Describe the outcome in a Syrus chat — a feature, a refactor, or a larger initiative. Syrus reads the selected repositories and drafts the plan. Nothing becomes work until you confirm it.
Big goals become epics: dependency-ordered stacks of scoped tickets — “Jobs”, in Syrus — that can be accepted, edited, or rejected before any code is written.
A developer releases the plan for execution — or triggers Syrus directly with a label on a GitHub issue, right where your code already lives.
Clones, branches, and implements in an isolated workspace — many changes in flight at once, even on the same files — then opens a pull request with the diff and a step-by-step test plan. The full transcript stays on the job record.
Your own checks — tests, lint, typecheck — grade every change, and a red result sends the agent back before the PR even opens. Syrus answers review comments; your developer brings the judgment machines can't.
Human approval gates the merge — self-review, two-person, or a designated final say: your policy. Once approved, Syrus lands it for you — rebased, re-checked on the exact branch, merged in dependency order. Branch protection, never bypassed.
Why Syrus
More output, real guardrails, and full visibility into what's being built — so you can move faster without gambling on quality.
Run many changes on one codebase at once — even changes that touch the same files. Every job works in its own isolated workspace, in-flight branches are rebased against each other automatically, and ordered work ships as stacked pull requests that land in sequence.
Your own checks — tests, lint, typecheck — grade every change, and a red result sends the agent back to fix it before a PR ever opens. Then human approval gates the merge, per the review policy you set, so generated changes stay inside the same review loop as human-written code.
After sign-off, the landing queue takes over: rebase, re-run your checks on the exact branch about to merge, then merge in dependency order. Epics land as one green unit or not at all — no half-merged features on your main branch.
Goals become tracked epics and tickets, and every attempt keeps its full transcript, diff, and review — an append-only record of who asked for what and exactly what the agent did. Operational dashboards track queues, workers, and host pressure, while repository throughput metrics show PR creation, output, landing, and review flow without hiding sparse samples.
Every run records what it spent. A spending dashboard breaks AI cost down by epic, user, and repository, so operators can understand model usage alongside the work that produced it.
Self-hosted on your infrastructure, and your code goes only to the plugin-backed model provider you choose, on your own credentials. The bundled provider adapters are Claude Code, Codex, Antigravity, and Muse Code. Syrus only ever calls out: no inbound webhooks, no public endpoint to expose. Your issues, reviews, and merges stay in GitHub.
Seven ways in, one execution model
A ticket, a narrated screen recording, a review comment, a failing check, a scheduled chore — each becomes, or updates, the same tracked job, with every attempt on one timeline. Nothing happens off the books.
A labeled GitHub issue becomes a tracked Job, a branch, and a pull request on the next repository poll.
Your review comment triggers a follow-up commit on the same branch; teammates' comments queue for one-click apply.
A failing check kicks off a bounded, automatic repair attempt.
Record your screen with narration from chat. Syrus extracts issues, desired behavior, and screenshots before the agent proposes the work.
A confirmed chat proposal or a direct prompt, no ticket needed — chores, experiments, urgent work.
Cron or one-shot runs for dependency chores, docs sweeps, and repository hygiene.
An unmergeable branch gets a deterministic rebase — the agent steps in only if judgment is needed.
Download the desktop app: it sets up a complete local Syrus with Docker, or connects to your own instance. The public docs and GitHub repo are the starting point for installation, operations, bug reports, and community contributions.
Open-source onboarding
Syrus is an open-source, self-hosted project. Everything you need to evaluate it starts with the release artifacts, documentation, and public issue tracker.
No account signup, vendor workspace, or product gate: fork it, install it, inspect it, and run it against repositories you control.