We build knowledge infrastructure.
Our tools turn scattered documentation into queryable, graph-connected knowledge that AI applications can reason over.
Conduit combines vector search with knowledge graph traversal. When it finds a relevant concept, it walks relationships to discover connected knowledge the query didn't mention. That's the difference between keyword-matching and understanding.
pip install 'conduit-ai[local]'from conduit_ai import LocalConduit
conduit = LocalConduit("./my-kb")
conduit.install_pack("snowflake-2026.04.ckp")
results = conduit.search("How does Cortex Search work?")| conduit | The engine — self-host with Docker Compose, or embed in Python |
| conduit-py | Python SDK — LangChain retriever, embedded DuckDB engine, CLI |
| conduit-mcp | MCP server — plug into Claude Code, Cursor, Pi, or any MCP agent |
| knowledge-packs | Seed packs — Snowflake, AWS, Databricks, GenAI (14,944 zettels) |
- Data engineer in a notebook? →
pip install 'conduit-ai[local]'+ install a knowledge pack - Building a chatbot/assistant? → Self-host the engine with Docker Compose
- Using Claude Code or Cursor? → Install conduit-mcp for knowledge graph access in your agent
- Want to try it first? →
pip install conduit-aiand runconduit inspect snowflake-2026.04.ckp