Python guide

The Python track. pip install kglite, then import kglite. This is the headline distribution path — the wheel ships a compiled extension (PyO3 wrapper over the pure-Rust kglite engine) and the kglite-mcp-server command (the same pure-Rust MCP server, bundled into the wheel as of 0.10.26 — no separate install). A standalone cargo install kglite-mcp-server binary is also available; see MCP Servers.

Coming from 0.13? The code-graph builder and dataset loaders moved out of the wheel in 0.14 — see the 0.13 → 0.14 migration guide. Pin back anytime with pip install "kglite<0.14".

If you’re embedding the engine directly in a Rust binary, the Rust guide is for you.

Start here

• Getting started — install, build your first graph, run a Cypher query, save / load a .kgl.

• Core concepts — nodes, relationships, storage modes, the selection model.

How-to guides

How-to guides

• Guides
• Data Loading
• Cypher Queries
• MCP Servers
• Authoring MCP skills
• Durable embedded apps
• Blueprints
• Querying (Fluent API)
• Traversal Hierarchy
• Semantic Search
• Spatial Operations
• Timeseries
• Graph Algorithms
• Import and Export
• Using with AI Agents
• OKF Ingestion
• Common Recipes

Python-specific topics

• Transactions — begin() / commit() / rollback(), snapshot isolation, OCC.

• Error handling — typed exception hierarchy (KgError + 16 subclasses).

• Value projection — NULL handling, CASE branches, optional property semantics.

• Platform and artifact support — runtime-tested, release-built, and best-effort targets; wheel and source-build policy.

Migrations

• Neo4j → KGLite — evaluate or adopt KGLite from an existing Neo4j database / driver code.

• 0.13 → 0.14 — the handover release: code-graph building moved to codingest, dataset loaders to kglite-datasets. Escape hatches and per-surface migration table.

• MCP 0.6 → 0.9 — older MCP server users.

• MCP pre-0.9.20 — the bundled-binary → Python-implementation switch.