Cognition Labs launched a Code Graph Indexer for its autonomous engineer Devin. By maintaining incremental AST dependency topologies across repositories, Devin reduces symbol and call-chain retrieval times in 1M+ line monorepos from 8s to under 600ms, while cutting irrelevant context token overhead by 65%.
Key Takeaways
- βIncremental AST and dependency graph topology drastically speeds up agent comprehension in huge repos.
- βSymbol retrieval in million-line monorepos drops to 600ms with a 65% token saving.
- βLive for all Devin Enterprise users with support for self-hosted code repositories.