trueforge

Why we included this project

Most agent frameworks treat a single model call as the whole job; TrueForge supplies the execution layer that sits between the LLM and a finished task. It runs the plan-act-observe loop end to end, routing tool calls over MCP, running generated code inside a sandbox, and keeping context and memory under control, so an agent can finish multi-step work instead of stalling after one reply. For anything that touches real systems, built-in human-in-the-loop approvals, per-run observability, and governance hooks give an agent room to act on sensitive actions without free rein. The project is a monorepo with a dedicated SDK and UI around the core, so you can embed it as a thin library in a local prototype or scale it into a distributed, multi-replica deployment. Engineers who want the agent to actually complete the work, not just loop the model again, get a working runtime here.