GitHub radar
Open Source Agent Runtime by TrueFoundry
TrueForge is an open source agent wrapper created by TrueFoundry that makes any LLM an agent with isolated code execution, MCP integration, and human approval.
TrueForge is an open source agent wrapper developed by TrueFoundry that wraps around any LLM to make it into an agent. It was released about a month ago and now has 3,400 stars. It provides model calling functionality, tool invocation, isolated code execution (using Daytona), session management, and human approval (where the agent stops and asks for permission before performing important actions while keeping track of all its actions in an audit trail). It works with OpenAI, Anthropic, Google Gemini, and other OpenAI compatible endpoints. It supports skills (instruction sets stored in Git) which can be added to the agent without modifying the underlying runtime and can be updated dynamically. There are three ways to interact with it: through a chat UI, via an HTTP API for integrating it into your applications, or using embeddable UI elements to build your own UIs.
Why a vibe-coder should care
Rather than developing your own agent infrastructure, you can have a production ready runtime up and running in just a few minutes. With human approval, you know that your agent won’t perform important actions autonomously without your consent. Skills allow you to dynamically change what your agent is capable of without having to redeploy it.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Deploy TrueForge from here: https://github.com/truefoundry/trueforge — follow the README, run via `npx @truefoundry/trueforge`, open the local interface and show it to me. Ask me if you need a model API key.
Runs on a regular laptop — Node.js 22.14 and a language model key required (OpenAI, Anthropic, or Google).
Open on GitHub▌ More finds