Sourcebot
Self-hosted codebase search and Q&A for humans and AI agents.
About
Sourcebot is a self-hosted tool for understanding large codebases. It maintains an up-to-date index of thousands of repositories across GitHub, GitLab, Bitbucket, Gerrit, and other hosts, so you can search and navigate code without checking anything out locally — particularly useful when ramping up on unfamiliar services or tracing behavior across microservice repos.
Search is built on Zoekt trigram indexes with universal-ctags symbols: multi-repo regex, symbol, language, file, repository, and branch-filtered queries, plus a file browser, Git history, and blame. On top of search sits Ask Sourcebot, which exposes the search, code-navigation, and file-reading APIs to an LLM as tool calls and answers questions with inline citations — no vector embeddings, just tool-use over a solid index. You bring your own API key to any of 11+ supported model providers, and agents get an MCP endpoint at /api/mcp as a code-context layer.
It is self-hosted and free to use under an MIT license, which is the appeal and the cost: you run the Postgres/Redis-backed infrastructure, and the MCP agent layer plus named search contexts sit behind a paid tier. A public demo runs at app.sourcebot.dev if you want to try before deploying.
What it can do
- Multi-repo code search: regex, symbol, language, file, repository, branch filters
- Zoekt trigram indexes + universal-ctags symbol navigation; Git history and blame
- Ask Sourcebot: LLM Q&A over your code with inline citations, BYOK to 11+ providers
- MCP endpoint (/api/mcp) exposing code context to AI agents
- Self-hosted and MIT licensed; data never leaves your infrastructure
Setup
- Deploy with Docker Compose following docs.sourcebot.dev, or try the public demo at app.sourcebot.dev.
- Connect your code hosts (GitHub, GitLab, Bitbucket, Gerrit) via OAuth or tokens.
- Configure environment secrets (AUTH_SECRET, SOURCEBOT_ENCRYPTION_KEY) in your deployment.
- Add your model provider API key (BYOK) to enable Ask Sourcebot; expose /api/mcp to agents on paid tiers.
Prerequisites
Docker (or Kubernetes) and a server you control; OAuth app or access tokens for your code hosts; an LLM provider API key for the Q&A features.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: 3,990 GitHub stars (sourcebot-dev/sourcebot); MIT license confirmed via GitHub API; Self-hosted codebase search: Zoekt trigram indexes, universal-ctags symbols, multi-repo regex/symbol/language/branch search; Ask Sourcebot Q&A with citations; BYOK to 11+ model providers — verified against docs and 2026 integration research; MCP endpoint (/api/mcp) for agents is a paid-tier entitlement — verified via third-party integration docs. Manual review: smaller star count reflects niche, not weakness; self-hosting cost (Postgres/Redis infra) is the honest tradeoff; v5.x actively maintained. License: MIT.
★ 3,990 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 3,990 GitHub stars (sourcebot-dev/sourcebot); MIT license confirmed via GitHub API
- · Self-hosted codebase search: Zoekt trigram indexes, universal-ctags symbols, multi-repo regex/symbol/language/branch search; Ask Sourcebot Q&A with citations; BYOK to 11+ model providers — verified against docs and 2026 integration research
- · MCP endpoint (/api/mcp) for agents is a paid-tier entitlement — verified via third-party integration docs
- · manual review: smaller star count reflects niche, not weakness; self-hosting cost (Postgres/Redis infra) is the honest tradeoff; v5.x actively maintained
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.