Finding a bug is half the job.
The expensive half is fixing it: correctly, cheaply, without breaking three other things.
When validation fails, an agent diagnoses, picks the cheapest fix (source-only → graft → watcher-only → full reindex), redeploys, and re-tests.
Then a
The modular data layer powering on-chain AI.
Token not launched yet.
Incubated & Backed by @ChainGPT_Labs
MVB9 @BNBCHAIN @yzilabs
PostWeb @outlierventures
Joined January 2019
- Dirty secret of blockchain indexing: Most of it is wrong somewhere, and nobody knows. An indexer can sync, look healthy, and still compute the wrong number for a slice of users. "Looks healthy" isn't good enough. 👇
- A reverse-engineered spec isn't data. Someone still has to write the indexer. We don't. The machine writes the subgraph, schema, and WASM watcher, then compiles it for real. If it fails, it reads the error and rewrites. Up to 3× per stage. Only code that builds moves forward.
- DeFi ships new protocols every day, often unannounced. Most teams discover them with analysts scrolling explorers. We run a machine that never sleeps: → sweeps 20+ chains → spots on-chain fingerprints of new DeFi → finds a real wallet with a live position → feeds it
- Integrating a DeFi protocol used to mean a senior engineer reading bytecode for a day. We reduced it to one input: One real user's position. At one block. From that, the engine reverse-engineers a formula proven against the chain to 0.5%, then that formula works for every

