TrueFoundry's TrueForge: Cheaper Open-Source AI Agent Tasks

San Francisco-based TrueFoundry, a machine learning startup co-founded by former Meta engineers, has introduced TrueForge, an open-source AI agent harness. Released on Github under the permissive MIT License, TrueForge aims to significantly cut the cost of AI agent task completion for enterprises, claiming savings of 30% to 75% compared to Anthropic's Claude Managed Agents. This new offering seeks to provide developers with greater control over AI agents and tools while addressing the growing expense associated with their proliferation.

Significant Cost Savings Unveiled

TrueFoundry detailed its findings in a recent blog post, showcasing TrueForge's efficiency. When paired with the open-source GLM-5.2 LLM, TrueForge successfully completed 11 out of 14 tasks on DevRev’s Enterprise-Bench, a testbed for multi-step tool use across CRM, issue tracking, and document management systems. This configuration achieved the tasks at a cost of just $2.90, a remarkable 75% reduction compared to the $11.80 required by Anthropic's Claude Managed Agents running Claude Opus 4.8 for the same outcome.

Even when running the identical high-performance Opus 4.8 model in both harnesses, TrueFoundry asserts a substantial 30% cost saving with TrueForge, costing $8.50 versus Claude Managed Agents' $11.80. These figures underscore TrueForge's ability to optimize operational expenses for AI agent deployments, regardless of the underlying model choice.

Strategic Vision: Open Source and the AI Gateway

TrueFoundry's decision to release such an efficient harness for free aligns with their broader business strategy. Anuraag Gutgutia, TrueFoundry’s co-founder and COO, explained that customers sought a vendor-neutral way to launch managed agents at a lower cost. TrueForge is designed to complement existing solutions, providing an open-source alternative that integrates seamlessly with diverse AI models.

The company already offers a paid "AI Gateway" that provides a centralized control layer for enterprises, managing model access, credentials, permissions, budgets, and observability. TrueForge operates "above" this gateway, handling the iterative process of an agent reasoning, calling tools, receiving results, and continuing tasks. TrueFoundry envisions its AI Gateway as the fundamental layer beneath all agents and harnesses, channeling traffic for unified governance and control.

Technical Innovation for Cost Efficiency

TrueForge's architecture is specifically engineered to minimize waste, particularly through advanced context engineering. It intelligently manages how much information is fed back into the model during each step of an agent's operation. This includes deferring the loading of tool schemas until necessary, offloading large tool results to files, processing structured data via code, and automatically compacting lengthy conversations, with a default threshold of 50,000 tokens.

Furthermore, TrueForge adopts a unique "sandbox-as-a-tool" approach. Unlike runtimes that keep an agent perpetually isolated, TrueForge's core agent loop resides on the server, provisioning a sandbox only when code execution or file interaction is required. This method significantly reduces unnecessary compute resources, allowing servers to run more agents concurrently and directly contributing to lower model spending.

Position in the Agent Harness Ecosystem

In the burgeoning field of AI agent orchestration, TrueForge enters alongside established and emerging players. Competing open-source options include DeepSeek Harness, OpenAI Codex CLI (coding-focused), and LangChain Deep Agents, all offering varying degrees of modularity, model flexibility, and deployment options. Anthropic's Claude Managed Agents, by contrast, offers a fully managed, proprietary runtime with tight model coupling.

TrueForge distinguishes itself by combining an open-source, vendor-neutral harness with a clear and scalable path from local development to shared production environments using Docker Compose or Helm. Its key differentiator is this fusion with an optional enterprise governance plane provided by TrueFoundry's commercial AI Gateway, offering both flexibility and robust control. Beta users like NetApp have leveraged TrueForge for applications such as incident response and self-service tools, demonstrating its practical enterprise utility.

TrueFoundry's Growth Trajectory

Founded in 2021, TrueFoundry initially focused on MLOps, helping enterprises deploy and operate machine learning models on Kubernetes. This foundation evolved to managing LLM applications and now agents, reflecting the broader industry shift. Backed by approximately $21 million in funding, including a $19 million Series A in February 2025 led by Intel Capital, the company has expanded its reach and offerings.

By 2025, the AI Gateway became central to its business, handling routing, authentication, and budget management for large-scale enterprise AI traffic. The company, which acquired UK-based MLOps vendor Seldon AI in June 2026, processes over 10 billion requests per month through its gateway for more than 30 paid customers, including Fortune 1000 organizations like Siemens Healthineers and Synopsys. TrueForge thus represents an extension of this comprehensive strategy, providing an open-source runtime that can be governed by TrueFoundry's overarching platform.

FAQ

Q: What is the primary benefit of TrueFoundry's TrueForge for enterprises?

A: TrueForge's primary benefit is its ability to significantly reduce the cost of running AI agent tasks, claiming 30% to 75% savings compared to Claude Managed Agents, while also offering greater developer control and vendor neutrality through its open-source license.

Q: How does TrueFoundry plan to monetize TrueForge if it's open source and free?

A: TrueFoundry offers TrueForge as a free, open-source harness to drive adoption and traffic towards its commercial "AI Gateway." This gateway provides a crucial governance layer for enterprises, offering centralized control over models, tools, access policies, budgets, and observability, thereby becoming the common control plane for diverse AI workloads, including those managed by TrueForge.

Q: What distinguishes TrueForge from other open-source AI agent harnesses?

A: TrueForge's key differentiator is its combination of an open-source, vendor-neutral harness designed for easy progression from local development to shared production, alongside the optional, comprehensive enterprise governance and control provided by TrueFoundry's commercial AI Gateway. It also features innovative context management and a "sandbox-as-a-tool" architecture to optimize costs.