TrueFoundry releases open-source AI agent harness TrueForge
Technology firm TrueFoundry has launched its open-source TrueForge harness, aiming to reduce the cost of AI agent task completion.
San Francisco-based machine learning startup TrueFoundry has released its open-source TrueForge harness. The tool is designed to manage AI agent operations and enhance developer control while aiming to significantly reduce costs compared to competing solutions.
TrueFoundry claims that TrueForge can complete tasks 30%-75% cheaper than when using Anthropic's Claude Managed Agents. The company tested TrueForge in conjunction with the open-source GLM-5.2 language model, completing 11 out of 14 tasks in the DevRev Enterprise-Bench test. The reported cost was 75% lower than with Claude Managed Agents.
"Our customers have been asking for tools to launch managed agents," said Anuraag Gutgutia, co-founder and COO at TrueFoundry. "TrueForge provides this capability in a way that allows for vendor-neutral use and lower costs." The company also offers a paid AI Gateway service for enterprises, focusing on model and access credential management.
TrueForge focuses on context engineering to minimize unnecessary data transfer to models. It delays loading tool schemas, delegates tasks to sub-agents, and moves large results into files. The company believes these optimizations directly reduce model usage costs and allow more agents to run concurrently on a server.