murrmurrplatform/opensweep — explained in plain English
Analysis updated 2026-05-18
Automatically sweep a codebase to find bugs, stale docs, and missing tests.
Let an agent implement a ticket and open a pull request that fixes itself until it passes review.
Chat live with an AI agent inside a sandboxed clone of your repo to ask questions about the code.
Self-host an autonomous coding agent dashboard instead of paying for a managed service.
| murrmurrplatform/opensweep | 0marildo/imago | 0xdfi/glm-5.2-1m-4x-dgx-spark | |
|---|---|---|---|
| Stars | 3 | 3 | 3 |
| Language | Python | Python | Python |
| Setup difficulty | — | easy | hard |
| Complexity | — | 2/5 | 5/5 |
| Audience | developer | general | researcher |
Figures from each repo's GitHub metadata at analysis time.
OpenSweep runs a set of AI agents whose job is to keep a codebase healthy without much hands-on effort from a human developer. You point it at your repositories and the agents sweep through the code, write documentation that stays up to date, spot bugs and gaps before anyone else notices them, and then pick up that work themselves: they implement fixes, open pull requests, review their own changes, and keep iterating until the pull request is ready to merge. The person using OpenSweep is left with just two jobs: approving tickets and merging pull requests. The system works through two loops running side by side. The first is discovery, where agents scan the repo, build a tree of living documentation, and file findings such as bugs, missing tests, or documentation that has gone stale, each one stamped with when it was last checked. The second loop is delivery, where a finding gets turned into a ticket, approved by a human, then implemented by an agent as a draft pull request, checked by a separate review agent, and cycled through fixes until it converges into something mergeable. There is also a chat feature called Ask, letting you talk directly with an agent inside a sandboxed copy of any repo. OpenSweep is self-hosted using Docker, needs about 4 gigabytes of free memory, and comes with a simple start script that sets everything up, including a bundled login system. It works with several AI providers, including Claude Code, OpenAI Codex, or fully local models, so you are not locked into one company's AI. On the security side, agents never touch your actual codebase directly. They work inside disposable clones fetched from GitHub, get only a limited set of environment variables, and use scoped, temporary access tokens rather than your real credentials. Webhooks are verified and encrypted where configured. Self-hosting OpenSweep is free forever under the Elastic License 2.0, meaning you can use and modify it but cannot resell it as a hosted service to others. A paid, fully managed cloud version is also offered for people who do not want to run the infrastructure themselves.
AI agents that scan your codebase, write documentation, file bug reports, and drive their own pull requests to a mergeable state.
Mainly Python. The stack also includes Python, Docker, Neo4j.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.