Blitzy Inc. raised $200 million in May, reaching a valuation of $1.4 billion. The company develops a platform that deploys thousands of autonomous coding agents to work on enterprise codebases in parallel.
The platform utilizes Neo4j graph databases to provide agents with system context. By reverse-engineering existing environments, Blitzy creates a dynamic graph of the codebase that integrates with systems like GitHub and GitLab.
According to engineering director Neeraj Deshmukh, traditional vector searches limit an agent's context to roughly 200 000 to 300 000 tokens. Graph structures allow agents to access specific dependencies without losing information in large-scale projects.
The company reported an 84.95% score on the SWE-Bench Pro benchmark in June. Quality control involves human approval of action plans and automated testing of all generated code, with secondary agents monitoring for potential hallucinations.
