Autoheal raises $7.9 million to improve software agents

Autoheal raises $7.9 million in seed funding for its platform that evaluates and repairs AI agents used in software development, according to SiliconANGLE. Innovation Endeavors led the round, with participation from Emergent Ventures, U&I Ventures, Darkmode Ventures, Batch Ventures and Param Hansa Values.

The platform runs within a customer's private cloud and connects with coding agents, code repositories, integration pipelines and observability tools. It builds a shared engineering context graph and uses an Evaluator agent to score other agents based on measures such as integration failures and incident reports.

A Healer agent can propose changes to model selection, prompts, tools and skills through pull requests. Each change is version controlled, tested against historical benchmarks and requires human approval. Autoheal plans to develop reinforcement learning methods that train agents on customers' private engineering data and support small language models running in private cloud environments.