Documents become a graph
Split source material into chunks, extract entities and facts, create summaries, and connect related passages.
KNOWLEDGE GRAPHS / GITHUB
Give your documents connections worth retrieving.
Build a searchable knowledge graph where proposed facts and relationships are checked before they enter the graph.
THE PROJECT, IN PRACTICE
What you can do with it.
A proposal and a check
Claude Code or Codex drafts the text. Jev answers typed questions that gate what is accepted.
Retrieve beyond similar words
Vector search finds relevant passages; graph links lead to related context. The MCP server returns source citations to assistants.
Keep the evidence trail
Passages, facts, and relationships trace back to the source material and the decisions that accepted them.
FROM IDEA TO USE
How it works.
- 01
Choose the corpus
Point the builder at supported documents or folders, and review the proposed ontology.
- 02
Build and audit
Extract, embed, merge, link, and audit through the staged pipeline.
- 03
Put the graph to work
Query it through MCP or export Q&A and fact sets for model training.
pip install jev-graph-builder
# Configure your TypeSafe key in your environment
# Start a build with Claude Code
jev-graph-builder --harness claude_code build ./my-docsRequires Python 3.11+, a configured System One service, an authenticated coding agent, and PostgreSQL with pgvector or a supported container runtime.
TAKE IT FOR A SPIN
Start here.
Install the package, configure the decision-model service, and select a document folder. The builder pauses for approval of its initial ontology; interrupted builds can be resumed.
STATUS & REQUIREMENTS
The default decision service is TypeSafe’s Jev. Other System One endpoints can be configured. Model and provider usage are separate from the local graph tooling.