Tiny devlog from the code cave:
I’m building CodeMeridian, a local code knowledge graph for AI coding tools.
The idea came from a very specific pain: AI agents feel great on small repos, but once a project grows into MVP territory, they begin guessing from nearby files, stale docs, and half-remembered architecture rules.
CodeMeridian indexes C# and TypeScript/TSX into Neo4j, then exposes MCP tools so the agent can ask things before editing:
What calls this?
What tests cover it?
Is this graph stale?
What files are actually in scope?
It does not replace Copilot, Claude, Codex, or local models.
It is more like giving them a project map with a red line drawn around the dangerous parts.
Repo:
A persistent code knowledge graph that gives GitHub Copilot a grounded, structural understanding of your codebase. It acts as the **determin











