Summary
On August 19, 2026, TrueFoundry launched TrueForge, an open-source (MIT) vendor-neutral AI agent harness on GitHub and PyPI, positioned as a cheaper alternative to Claude Managed Agents (30-75% lower task cost). It runs the model-to-agent execution loop with 40+ built-in tools, MCP integration, sandboxed code execution via Daytona, human-in-the-loop approvals, context compaction, and per-task model swapping, exposed through a chat interface, an HTTP API with a TypeScript SDK, and an embeddable UI.
What changed
TrueFoundry released TrueForge under the MIT license on GitHub and PyPI: an agent execution harness with 40+ built-in tools, MCP tool integration, sandboxed execution via Daytona, human-approval checkpoints, automatic context compaction, generative UI streaming, Tavily web search, and per-task model swapping; free with your own API keys, with a usage-priced hosted option.
Why it matters
An open, vendor-neutral harness that swaps models per task and undercuts managed offerings on cost pressures the 'managed agent runtime' business (Anthropic's Claude Managed Agents, OpenAI's AgentKit) and pushes agent orchestration toward commodity, self-hostable infrastructure.
Evidence excerpt
"TrueFoundry's open source AI agent harness TrueForge boasts 30%-75% cheaper task completion than Claude Managed Agents." (VentureBeat)