Now open source · MIT License

Your codebase,
finally understood

rekipedia turns any repository into a searchable wiki with grounded Q&A, exact file:line citations, and an MCP server so your AI agent actually knows your code.

pip install rekipedia → View on GitHub
~/my-project
reki scan .
✔ Scanned 312 files · 4,891 symbols indexed
✔ Wiki generated · .rekipedia/wiki/
Done in 4.2s
reki ask "how does the auth flow work?"
The entry point is src/auth/middleware.py:42
AuthManager.verify() validates the JWT and attaches
the user to the request context.
Sources: src/auth/middleware.py:42, src/models/user.py:18
Everything you need to understand any codebase

From a first-time scan to live AI agent integration — rekipedia covers the full workflow.

🔍
reki ask
Grounded Q&A
Ask anything about your codebase and get answers with exact file:line citations. No hallucinations — every answer is sourced from your index.
🌐
reki serve
Local Wiki UI
Spin up a browseable wiki at localhost:7070. Auto-generated architecture docs, module maps, and relationship graphs — all from your actual code.
🔌
reki mcp
MCP Server
Expose your knowledge base as an MCP server. Claude Code, Cursor, and any MCP-compatible agent can query your codebase with structured tools — not just file dumps.
🔥
reki hotspots
Architectural Risk
Find hub nodes (files with 40+ dependents) and bridge nodes (the connective tissue between clusters) before you start a refactor. Know what you're touching.
🚀
reki onboard
Instant Onboarding
Generate a structured architecture walkthrough for new team members in seconds. No more "read the code and figure it out" onboarding.
reki watch
Always Fresh
Auto-reindex on every file save. Your wiki stays in sync as you code — no manual rebuilds, no stale docs.

From repo to knowledge base in three steps
1

Scan your repo

rekipedia parses every file into a SQLite knowledge store — symbols, call graphs, module relationships, and dependency trees.

reki scan .
2

Ask anything

Query your codebase in plain English. Every answer comes with exact file:line citations — grounded, verifiable, no guessing.

reki ask "where is rate limiting handled?"
3

Connect your agent

Start the MCP server and connect Claude Code or Cursor. Your AI agent now queries a real knowledge base — not just whatever you remember to paste in.

reki mcp
Give your AI agent a real knowledge base

Most "AI + codebase" tools work by dumping file contents into context windows. That doesn't scale, doesn't persist, and doesn't let agents reason about structure.

With reki mcp, your agent can ask "If I change engine.py, what else breaks?" and get a real dependency graph back. Drop a .mcp.json in your project and Claude Code or Cursor connects automatically.

Claude Code Cursor Any MCP Client
askGrounded Q&A with file:line citations
search_nodesFind symbols, functions, and classes by name or pattern
get_contextFull context for any symbol — callers, callees, docstring
get_relationshipsDependency graph between modules and files
get_hub_nodesMost connected files — the load-bearing walls of your repo
get_impactBlast radius analysis — what breaks if you change X?

Get started in seconds
$ pip install rekipedia
$ uvx rekipedia init . && uvx rekipedia scan .
$ npx rekipedia init . && npx rekipedia scan .
$ pip install "rekipedia[rag]"

Works with any OpenAI-compatible endpoint. Run local with Ollama or connect to any cloud provider.

Ollama (default)
Run fully local, zero cost
REKIPEDIA_MODEL=ollama/llama4
OpenAI
GPT-4o, o1, and more
OPENAI_API_KEY=sk-...
Anthropic
Claude 3.5 Sonnet, Opus
REKIPEDIA_MODEL=claude-3-5-sonnet-...
Google Gemini
Gemini 2.0 Flash and Pro
REKIPEDIA_MODEL=gemini/gemini-2.0-flash
OpenRouter
100+ models via one key
REKIPEDIA_MODEL=openrouter/...
Azure OpenAI
Enterprise deployments
REKIPEDIA_MODEL=azure/gpt-4o

Stop copy-pasting code.
Start asking questions.

rekipedia is MIT licensed, language-agnostic, and works anywhere Python runs.

⭐ Star on GitHub View install options