Knowledge Tree
where knowledge takes root
Knowledge as an obsidian vault stored as markdown in git and ingested and indexed context with skills and tools for human and ai coordination. For details on how this works from a high level, read architecture
branches.yaml is the routing file for the knowledge tree and controls building the tree at compilation with controls for encryption, pulling from remote sources such as books, etc...
- indexing
- tagging
Source code used for knowledge management. Commands to interact with the knowledge bank exist in makefile.
Run make help to see the available commands.
One image, ghcr.io/teaglebuilt/kb, runs every mode:
| Mode | Command | Purpose |
|---|---|---|
kb-mcp-read |
python -m kb.mcp |
Read MCP: kb_search, kb_get_document |
kb-api |
python -m kb.api |
FastAPI /v1 + /openapi.json, and kb-mcp-write at /mcp |
| sync | python -m kb sync |
LanceDB → Qdrant |
- docs/api.md - canonical HTTP contract, auth, configuration and the migration from the old names.
- docs/mcp.md - the two MCP servers and their exact tool inventories.
- docs/write-plane.md - why
kb-apiexists and how writes land in Qdrant. - docs/read-plane.md - how documents are read back, including OpenWebUI.
Automated sources (n8n YouTube ingestion) POST /v1/documents to kb-api
instead of running the local pipeline. Local commands: make mcp (read MCP,
stdio), make api (kb-api on :3000; ANON=1 for anonymous local writes),
make openapi (print the OpenAPI document), make helm-deploy-api.