TrueFoundry, a San Francisco startup founded by former Meta engineers, released TrueForge, an open source AI agent harness licensed under MIT. The tool lets developers build and manage AI agents with greater control over models and tools while cutting costs significantly.
TrueForge claims 30-75% lower expenses for task completion compared to Claude Managed Agents. The framework works with any AI model, avoiding vendor lock-in. Developers can fork it, modify it, self-host it, and incorporate it into commercial applications.
The release targets a real pain point in enterprise AI. As companies deploy more agents, they struggle with cost control and flexibility. Most managed agent platforms charge premium rates for hosting and API calls. TrueForge strips away that overhead by running on developer infrastructure.
The harness provides standard features for agent systems: tool orchestration, prompt management, and logging. Developers retain full visibility into how their agents operate and can optimize prompts and tools without platform constraints. Self-hosting means no recurring vendor fees for simple agent tasks.
TrueFoundry's background in machine learning infrastructure gives the team credibility here. The company previously built tools for ML operations and model management, so they understand enterprise deployment challenges firsthand.
The open source approach matters. Closed platforms lock companies into specific models and pricing. TrueForge lets enterprises use cheaper models from OpenAI, Anthropic, or open source providers like Llama without paying a markup for the framework itself.
Competition in agent frameworks is intensifying. Anthropic offers its managed agents solution, while frameworks like LangChain and LlamaIndex serve similar needs. TrueForge differentiates on cost and control rather than managed convenience.
The cost advantage comes from avoiding intermediary markup and enabling cheaper model selection. A task that costs $100 on Claude Managed Agents might cost