Sabitlenmiş Tweet
I left Amazon to build the one thing every AI coding agent secretly needs… and nobody was shipping.
You know the exact moment it happens:
You’re in Claude Code / Cursor / Windsurf.
Your agent has been crushing it for 20 minutes.
Then it confidently deletes the wrong function, breaks three downstream services, and swears the call graph looks “fine.”
Because it doesn’t *actually* know your codebase.
It’s just hallucinating from thousands of tokens of chat history that just got compacted again.
I watched this exact failure mode every single week over the last two years while scaling agentic workflows.
So I quit.
I combined the compilation and static analysis techniques I learned at Georgia Tech with @abewolke's graph-driven vision and built the missing layer.
Then I withdrew.
Introducing:
Supermodel Tools: the first public Code Graph API for AI agents.
It doesn’t “remember” snippets from your chat.
It statically analyzes your entire repo and gives your agent a persistent, queryable world model:
* Full call graphs
* Dependency maps
* Domain classifications
* Program dependence graphs
All in sub-second queries. Survives compactions. Survives agent switches. Survives you closing the tab and coming back tomorrow.
Your agents finally stop guessing.
They navigate your code like they actually live there.
We’re launching today.
If you’re tired of watching billion-dollar models act like interns who forgot the org chart… this one’s for you.
Supermodel@supermodeltools
Introducing Supermodel: persistent Code Graphs for AI agents. Architectural knowledge that survives compactions and resets. → Sub-second graph queries → Zero data retention → MCP server for Cursor, Claude Code, Codex, Hermes, etc.
English
