Stacktown Platform
Mneme
A knowledge graph that captures the thinking behind your code. Intent, provenance, and documentation freshness — queryable by humans and AI agents.
AI Writes Code Faster Than Anyone Can Understand It
Institutional knowledge — the why behind code — has always been fragile. It lives in PR discussions, design docs, and conversations that fade as teams change. AI code generation makes this dramatically worse: AI produces code without leaving behind intent, provenance, or reasoning.
The gap between “code that exists” and “code that's understood” is widening faster than human processes can close it. Mneme captures the thinking that slips between the cracks — building a queryable knowledge graph that records not just what depends on what, but who generated it, why it was built, and whether the documentation is still accurate.
What Mneme Captures
Four layers of knowledge that traditional documentation misses — automated and queryable.
Queryable Knowledge Graph
Every code symbol is a node in a graph with structural relationships, provenance records, intent annotations, and documentation links. Query by function, author, generation session, or purpose.
Provenance Tracking
Know which AI session generated a function, who requested it, and when. As AI-generated code volume explodes, provenance is the difference between understood code and mysterious code.
Intent Annotations
Capture why code exists, not just what it does. Intent annotations persist the reasoning behind design decisions so they survive team changes and memory loss.
Self-Healing Documentation
Git post-commit hooks trigger automatic documentation updates when code changes. Mneme detects staleness, flags affected docs, and regenerates them — keeping documentation in sync with reality without manual effort.
Blast Radius Analysis
Query the dependency graph to see exactly what is affected when any function changes. Before you refactor, Mneme shows you the full blast radius — eliminating surprises from hidden dependencies.
AI Agent Integration
Designed to be queried by both humans and AI agents. AI coding assistants can query Mneme for context about existing code before generating new code — reducing hallucination and duplication.
Architecture
TypeScript Indexer
Parses codebases in TypeScript, JavaScript, Python, and Go. Extracts structural relationships, symbol definitions, and dependency graphs.
Go Graph Server
Owns the knowledge graph with Neo4j backing. Serves queries from humans and AI agents via CLI and API. Built for the scale of real codebases.
Local-First Deployment
All processing runs on your machine. No cloud required. Your code, your graph, your control. Cloud sync is optional, never mandatory.
Storage Abstraction
Neo4j Community Edition today, swappable tomorrow. The graph store sits behind an abstraction layer — change the backing store without changing the application.
Interested in Mneme?
Mneme is in active development. If you're dealing with AI-generated code at scale and losing institutional knowledge faster than you can document it, get in touch.