Technology

TrueFoundry Launches TrueForge Open-Source Agent Harness to Run LLMs

· GlobeWire Daily Tech Desk · Source: github.com

2 min read

Key takeawaysTrueFoundry has introduced TrueForge, an open-source agent harness designed to serve as a runtime layer that turns a Large Language Model (LLM) into a working agent. The software aims to handle the agent execution loop, covering model calls, Model Context Protocol (MCP) tools, skills, sandboxing, approvals, context management, and session state.
TrueFoundry Launches TrueForge Open-Source Agent Harness to Run LLMs

Image: GitHub · Photo source · Resized and letterboxed to 16:9; no AI generation

TrueFoundry has introduced TrueForge, an open-source agent harness designed to serve as a runtime layer that turns a Large Language Model (LLM) into a working agent. The software aims to handle the agent execution loop, covering model calls, Model Context Protocol (MCP) tools, skills, sandboxing, approvals, context management, and session state.

According to the repository documentation, TrueForge exposes its runtime capabilities through three distinct interfaces: a chat user interface, an HTTP API with a TypeScript SDK, and an embeddable UI SDK.

Core Features and Architecture

TrueForge provides infrastructure out of the box to help developers run agents reliably:

  • Initial Setup from Catalogs: Configure models, MCP servers, skills, and a sandbox once using customizable YAML catalogs.
  • Model Compatibility: Supports OpenAI, Anthropic, Google Gemini, and other catalog providers, alongside any OpenAI-compatible endpoint.
  • MCP Tools and Skills: Integrates remote MCP servers with header auth or OAuth and git-backed instruction packs loaded on demand.
  • Sandbox Execution: Offers isolated code and file execution via Daytona currently, with more providers planned, while keeping secrets inside the harness.
  • Human Checkpoints: Includes tool approval, user query prompts, and Generative UI in chat.
  • Context Engineering: Features subagents, deferred tool loading, Code Mode, large-result offloading, and compaction.

Deployment Options and Scaling

TrueForge scales across different environments:

  • Local Mode: Runs as a single process using SQLite for local machine testing. Documentation notes this mode has no default login and advises keeping it on localhost to prevent unauthorized access.
  • Hosted Mode: Utilizes Postgres and Redis, and can be deployed via Docker Compose, Helm, or Railway for shared or production environments.

Benchmarking and Contribution

TrueFoundry reports comparative benchmarks against Claude Managed Agents and deepagents within the repository's benchmark directory. TrueForge is released under the MIT License. Interested users can review the Quickstart guide and CONTRIBUTING.md file on GitHub to verify installation instructions, deployment methods, and contribution guidelines.

Share this article
Read the original source: truefoundry/trueforge: The open-source agent harness - the runtime layer that turns an LLM into a working agent. - GitHub

AI-assisted reporting; source claims are attributed and have not been independently verified.

About the author

GlobeWire Daily Tech Desk — AI-assisted news based on attributed published sources.

More to read