Skip to main content

Open Knowledge Tree

The source-driven knowledge engine.

Fetch a URL or DOI, and OKT extracts atomic facts, deduplicates them by meaning, links them into a concept graph, and accumulates syntheses you can search, cite, and reason over.

Get StartedUnderstand the System

The Data Pipeline at a Glance

Stage 1
Source
Fetch URL/DOI via provider chain
Stage 2
Facts
Decompose into self-contained atomic facts
Stage 3
Embed
Vectorize facts into Qdrant
Stage 4
Dedup
Merge idea-level duplicates by cosine distance
Stage 5
Concepts
Extract concepts, aliases, contexts -> graph
Stage 6
Summaries
Incremental per-concept synthesis slices
Stage 7
Synthesis
Authoritative definition per concept group
Phase 1
Research
search sources, fetch + process, open investigations
Phase 2
Query
search facts + concepts, read syntheses, walk related
Phase 3
Reports
author markdown, get back per-sentence fact annotations

Explore the Docs

Knowledge Flow

The 7-stage pipeline: source extraction, fact decomposition with coreference resolution, embedding-based dedup, concept graph, summaries, and synthesis.

MCP Tools

18 Model Context Protocol tools to fetch sources, search facts and concepts, track ingestion, and create annotated reports — all over OAuth 2.1.

REST API

HTTP endpoints for auth, repositories, sources, facts, concepts, investigations, reports, and providers. JWT and OAuth 2.1 bearer auth.

Local Dev

Run the full stack with Docker Compose and `just dev`. Postgres, Qdrant, FlareSolverr, and MinIO boot in seconds.

Architecture

Multi-database layout, Casbin RBAC, provider strategy, River task manager, Qdrant collections, and the full DB schema.