RepoSwarm workflow
RepoSwarm analyzes repositories and commits .arch.md architecture files to a shared hub that coding agents read for context.
- 279 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The service clones repositories and commits results to a shared hub.
- Repository code is sent to the chosen model provider for analysis.
Install
- Works in
Manual install
curl -fsSL https://raw.githubusercontent.com/reposwarm/reposwarm-cli/main/install.sh | shInstalls the `reposwarm` CLI. Then `reposwarm new --local` starts Temporal, the API, a worker and the UI. Read the script before running.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The service clones repositories and commits results to a shared hub.
- Repository code is sent to the chosen model provider for analysis.
This is third-party code. Review the repository files before installing.
What it does
RepoSwarm analyzes repositories and generates standard `.arch.md` architecture files. Results are committed to a shared hub that is added to an agent's workspace so it can see documentation across several repositories at once. The same files serve onboarding and architecture reviews. The pipeline: cache check, clone, type detection, structure analysis, prompt selection, AI analysis, storing results and cleanup. It supports GitHub, GitLab, AWS CodeCommit, Azure DevOps and Bitbucket and models from the Anthropic API, Amazon Bedrock or a LiteLLM proxy. It runs on Temporal workflows and locally needs Docker, Docker Compose and Git. Control it with the `reposwarm` command.
Who it is for. Architects and teams with many repositories where an agent does not know how neighboring services work and asks for explanations at every step.
Good fit when
- An agent edits a service and must know how it connects to neighboring repositories.
- A new developer needs an architecture overview of twenty repositories.
- Mentions of authentication must be found across all documentation.
- The architecture of two services should be compared with a diff.
Not a fit when
- The project is one repository and its own README is enough.
- Temporal and containers cannot be run.
- You need documentation with no human check, while AI findings can be wrong.
Example request
Add the repositories my-app and my-api, investigate both and find every mention of authentication in the generated files.Limitations
Generated findings need human review. Running Docker, Docker Compose and Git are required, plus Temporal, the API, a worker and the UI as services. The AI model is available through the Anthropic API, Bedrock or LiteLLM at your cost, and Bedrock needs configured AWS credentials. The CLI install recommends a `curl | sh` script, so read it before running. Caching uses DynamoDB or local files. The repository was previously named `loki-bedlam/repo-swarm`.
How to disable. Stop the services with the `reposwarm` CLI, delete the Docker containers and the hub folder. Remove the hub from the agent's workspace.
FAQ
What is the output?
`.arch.md` files with each repository's architecture, committed to a shared hub. The agent reads them as context.
How do I run an analysis?
Add a repository with `reposwarm repos add` and run `reposwarm investigate <name>`. Progress shows in `reposwarm wf progress` or `reposwarm dashboard`.
Can I process all repositories at once?
Yes, `reposwarm investigate --all --parallel 3` runs three in parallel. Use `reposwarm repos discover` to list them.
How do I search results?
With `reposwarm results search "authentication"`, and `reposwarm results diff` compares two repositories.
What if something fails?
Run `reposwarm doctor` for a health check, `reposwarm errors` shows stalls and failures and `reposwarm wf retry <id>` re-runs a workflow.
Discussion
No questions or reviews yet.
Related
WorkflowsGives the agent an understanding of your repository: it indexes the code and answers questions about it in Claude Code, Cursor, Codex, Windsurf and more
A skill that dissects an open-source project's architecture and returns one Markdown report with Mermaid diagrams and a comparison with rivals
An MCP server builds a graph of code entities, relations and flows so the agent finds the right files without reading everything
Git-native project context for agents: specs, decisions and rules live in Markdown beside the code and are read before edits
A CLI that analyzes a GitHub, GitLab or Bitbucket repository and builds a document kit for a new developer's first day.
A CLI that reviews code changes, fixes confirmed findings and repeats review before deterministic repository verification.
README in short
The README poses the multi-repository context question and covers quick start, the pipeline, common workflows, provider setup and the ecosystem. Apache-2.0 license.