Blitzy described its graph-based approach for autonomous coding agents that use Neo4j and Cypher and reported an 84.95% SWE-Bench Pro score
Feature update Provisional 62% confidence first seen
Blitzy said it is building autonomous coding agents that model enterprise codebases as knowledge graphs using Neo4j and Cypher, so the agents can understand dependencies before making changes. The company also reported achieving an 84.95% score on SWE-Bench Pro in June, positioning the approach as more reliable than broad search-based context.