Structural Code Parsing
A Tree-sitter turns your codebase into functions, classes, modules and their relationships. Not regex, not string matching — an actual AST.
Every tool call. Typed. Connected.
Traceable.
Observe every tool call. Extract typed knowledge. Build a knowledge graph that ages, contradicts, and traces back to source.
Knowledge that compounds via
Knowledge Graph
Everything your coding agent does, typed and connected. Decisions link to patterns, patterns link to code — the full history of your codebase, traceable to source.
Typed Knowledge
Decisions, patterns, conventions, architecture, discoveries, incidents — each thread is typed at extraction so your coding agent knows what kind of knowledge it's looking at. The type determines how it ages, how it's retrieved, and what it means in context.
Relationships
Your coding agent knows the auth refactor contradicts the session management decision. How? Because the threads are linked, not just stored. Track how decisions evolved, trace why a pattern was adopted, or surface when two threads contradict.
Multi-Signal Retrieval
This isn't nearest-neighbor lookup with a threshold. Multiple signals combine: vector similarity, keyword matching, type-aware boosting, confidence weighting, and relationship traversal. The retrieval pipeline surfaces what's relevant — not just what looks similar.
Observation Lineage
Every thread traces back to the tool call that produced it — the session, the file, the timestamp. Knowledge accumulates from the actual work, not the agent's narration of it. Repeated behavior across sessions becomes a named pattern, with receipts.
Code-aware with
CodeWeaver
Give your coding agent a mental model of your codebase. And best of all?
Your source code never leaves your computer.
Code Graph Navigation
Trace imports, find callers and understand what depends on what. Ask "what does changing this function affect?" before you change it.
Grounded Memories
Memories link to actual code entities — functions, classes, files and line ranges. A decision about your auth layer connects to your actual code.
Local-First Privacy
CodeWeaver runs locally and only stores structural metadata like names, types, line ranges and relationship edges in Neuroloom.
TypeScript and Python
Tree-sitter grammars for TypeScript and Python ship in the default install. The structural index is queryable from any MCP-compatible agent.
Integration
Built for Claude Code. MCP-compatible with Cursor, Windsurf, Augment Code and other agents. Programmatic access via SDK or REST API for custom workflows.
MCP-Native
Install the Claude Code plugin and Neuroloom starts working immediately - observations, context injection, and a knowledge graph out of the box. MCP-compatible with Cursor, Windsurf, Augment Code and other agents.
Python and TypeScript SDKs
Programmatic access for custom agent integrations. The same typed knowledge model and graph API, available without the MCP layer when you need direct control.
REST API
Every operation is available over HTTP. Bring your own client, build custom tooling or wire Neuroloom into existing agent pipelines. It's your knowledge graph, so have it your way.
Not what was stored. What's still true.
Start free. Knowledge graph and CodeWeaver included at every tier. No credit card required.