About Trueforge
TrueForge is an open source agent harness, the runtime layer that turns a language model into a reliable, long running agent. It is built by TrueFoundry and is designed to be dropped into an application, with support for MCP tools, skills, sandboxing, human approvals and subagents built in.
It has three parts. A core server runs the agent loop, which plans each turn, calls the model, executes tools and streams every step back. An HTTP API with OpenAPI documentation and a TypeScript SDK lets developers create and drive agents from code. A bundled chat interface with a React UI SDK lets users talk to agents in the browser and shape their configuration. Models, MCP servers and sandbox providers are bring your own, and sessions persist in SQLite or Postgres so conversations survive reconnects and restarts. Sandboxes are treated as a tool, provisioned only when an agent needs to run code, which lets one server run many agents at once. Harness features include context engineering through progressive skills, deferred tools, subagents and compaction, approval prompts before write or destructive tool calls, and generative UI.
It is aimed at developers building agent products who want the execution layer handled for them. The documentation lists no paid plans and no prices, and the software is open source.