Production signals in. Reviewed pull requests out.¶
Opsmith watches your logs, metrics and repositories, links what production shows to what the code shows, and ranks the result as engineering tasks. When a person approves a task, a sandboxed coding agent implements it and opens a pull request. After a person merges it, Opsmith checks that the problem is actually gone. It is self-hosted and works offline with built-in rules, with Claude, or with any OpenAI-compatible endpoint, including local models.
Two rules are never broken. Nothing changes code without a recorded human approval, and Opsmith cannot merge. A human always merges.
Status: stable, not yet proven in production. Read what has been tested and what is not protected before you point it at something valuable.
See it in a minute¶
You need uv and git. No API key and no Docker; once installed, the demo runs without network access.
git clone https://github.com/deadboyccc/Opsmith.git && cd Opsmith
uv sync --all-extras
uv run opsmith demo --dir ./opsmith-demo # ends with "final_status": "verified"
The demo builds a small repository with a real bug, feeds in logs and metrics, finds the problem from both sides, has a reviewer approve the fix, opens a pull request, merges it as a human would, and confirms the problem is gone. The overview shows it as a one-minute video; here it is in thirty seconds:
Your own reviews, on a schedule¶
Write a check in plain words, such as "audit this code for bugs and performance issues", and choose when it runs. A model carries it out, and every problem it can point to in the code becomes a task for review, under the same two rules. Checks need a model: the built-in offline rules cannot follow an instruction written in words. See Checks.
Where to go next¶
-
New here
- Overview: what it does and how it works
- Getting started: your own project, on your machine
- Concepts: leads, findings, tasks and their lifecycles
- UI tour: every page, with screenshots
-
Running it
- Configuration of projects, connectors and budgets
- Checks: your own review instructions, on a schedule
- Models: offline rules, any OpenAI-compatible endpoint, or Claude
- Deployment with Compose and the sandbox
- Operations: monitoring, upgrades, troubleshooting
- API reference
-
Trusting it
- Security: every guarantee and the test that proves it
- Security audit: what was found, fixed and left open
- Compatibility: what is verified, and how
-
Changing it