AGENT HARNESS

TrueForge Agent Harness for Windows — Free, MIT Licensed, 75% Lower Cost

TrueForge is an MIT-licensed runtime for building and operating AI agents with a developer’s choice of models, tools and infrastructure. It cuts agent loop costs by up to 75% versus Anthropic’s Claude Managed Agents while supporting any model via LiteLLM or native endpoints. The harness includes session persistence, sandboxing, MCP tools, context compression, and a TypeScript SDK for embedding in other products.

Download TrueForge Product overview

Why TrueForge beats Claude Managed Agents

MIT Licensed, Zero Harness Fee

TrueForge is free to use, modify, and redistribute. No per‑agent runtime fees — you pay only for the model vendor you choose (or run local open‑weight models for zero inference cost).

Model‑agnostic by Design

Bring your own model: GLM‑5.2, DeepSeek, Llama, Qwen, or any LiteLLM‑compatible endpoint. Switch providers without rewriting your agent loop.

Context Engineering Saves Tokens

Delays loading MCP tool schemas until needed, moves oversized tool results to files, and auto‑compacts long conversations at a 50k‑token threshold — cutting prompt size and inference spend.

Subagent Isolation for Parallel Work

Isolated tasks run in subagents, preventing context window bloat and enabling true multi‑agent orchestration on a single server.

Built‑in MCP & Tool Registry

Google, GitHub, Discord, Linear, and docs‑to‑skill presets come pre‑configured. Add custom tools via the registry without touching the core loop.

TypeScript SDK for Embedding

Ship a chat interface, HTTP API, or CLI inside your own product. The SDK handles auth, tool calls, and result streaming.

Self‑Host or Commercial Use

Fork the repo, run on your own hardware, or embed TrueForge in a SaaS offering. The MIT license permits commercial redistribution.

Proven on Enterprise Benchmarks

On DevRev’s Enterprise‑Bench (CRM, issue tracking, document management), TrueForge with GLM‑5.2 completed 11 of 14 tasks at $2.90 per run — 75% below Claude Managed Agents with Opus 4.8 at $11.80.

TrueForge vs Other Agent Harnesses

Feature	TrueForge	Claude Managed Agents	OpenCode	LangChain Deep Agents
License	MIT	Proprietary	MIT	MIT
Model Choice	Any via LiteLLM / native	Anthropic only	Any (local or API)	Any (LangChain ecosystem)
Hosting	Self‑host or embed	Fully managed (Anthropic)	Self‑host	Self‑host
MCP Support	Built‑in, delayed loading	Limited	Via plugins	Via LangChain
Context Compaction	Automatic, 50k‑token default	None	Manual	Manual
Cost per Run (GLM‑5.2)	$2.90	$11.80	~$3.50	~$4.00

TrueForge combines the permissive licensing of OpenCode with the enterprise‑grade context engine of LangChain Deep Agents, while undercutting managed‑agent pricing by a wide margin.

Getting Started with TrueForge

Download the server — Get the latest TrueForge release from fuseintelligence.org/downloads/trueforge/.
Choose your model — Point TrueForge at a local open‑weight model (GLM‑5.2, Llama 3) or an API endpoint via LiteLLM.
Configure tools — Enable MCP connectors for Google, GitHub, Discord, Linear, or add custom tools in tools.json.
Run your first agent — Use the TypeScript SDK or the built‑in chat interface to spawn an agent that can read files, run commands, and return structured results.
Monitor and optimize — View token usage, active subagents, and tool call logs in the dashboard; tune the compaction threshold for your workload.

FREE VIBEFUSE · FIRST WIDGET · VS CLOUD · CREATOR CHALLENGE · FOR AGENTS · START · BUILD IN PUBLIC · EARN · NO $100 PAYWALL

Join agentic developers who ship widgets, post on the forum, and sell on the marketplace — free VibeFuse license on every account. Create a free account · For Agents · Start · Build in Public · First Widget · vs Cloud Computers · Creator Challenge · Earn · Creator Playbook · FAQ.

Widget Wars: publish a free custom widget the agent can drive, open a forum build thread with install steps, earn Founders-track perks while seats remain — public rubric, no purchased votes.

✓ Free MIT license
✓ Any model via LiteLLM
✓ MCP tools built‑in
✓ Context compaction saves tokens
✓ TypeScript SDK for embedding

Try TrueForge free Agent harness guide Join free Marketplace Widget Wars Founders 50

Explore VibeFuse & harness guides

TrueForge FAQ

What is TrueForge?

TrueForge is an MIT-licensed agent harness runtime for building and operating AI agents with your choice of models, tools, and infrastructure. It targets lower agent-loop cost than managed offerings while remaining self-hostable and embeddable via a TypeScript SDK.

How does TrueForge cut costs versus Claude Managed Agents?

TrueForge charges no per-agent harness fee — you pay only the model vendor (or run local open-weight models). Features like deferred MCP schema loading, moving oversized tool results to files, and auto-compaction around a 50k-token threshold shrink prompt size and inference cost. Published claims cite up to ~75% lower agent-loop cost versus Anthropic's managed agent stack; verify on your own workload.

Which models does TrueForge support?

It is model-agnostic: GLM-5.2, DeepSeek, Llama, Qwen, Claude, GPT, and any LiteLLM-compatible endpoint. Switch providers without rewriting the agent loop.

Is TrueForge free for commercial use?

The MIT license allows use, modification, redistribution, and commercial embedding. You still pay whatever model or hosting costs you choose.

How does TrueForge relate to VibeFuse?

TrueForge is a harness runtime focused on cost-efficient agent loops; VibeFuse is a free Windows widget canvas that runs real agent CLIs as local PTY widgets with voice and an open marketplace. They solve adjacent layers — runtime economics versus desktop multi-agent workspace.