TrueFoundry Launches Open-Source AI Agent Foundation
US startup TrueFoundry has released TrueForge, an AI agent foundation software, as open-source under the MIT license. According to the company, it can reduce task completion costs by up to 75% compared to Anthropic's Claude Managed Agents. The design allows developers to start in local environments and transition directly to production, aiming to enable vendor-agnostic agent operations.

US startup TrueFoundry has released TrueForge, foundational software for operating AI agents, as open-source. Distributed under the MIT license, anyone can use it for free, and modification or integration into proprietary services is freely permitted. TrueForge is a mechanism called an 'agent harness' that repeats a cycle of 'think→invoke tool→receive result→think again' until the AI model achieves its goals.
TrueFoundry, founded in 2021 by engineers from former Meta (now Meta Platforms), is a San Francisco-based B2B machine learning startup. The company already provides AI Gateway, a paid service that allows enterprises to centralize access controls and cost management for AI models. TrueForge operates as the upper layer of that gateway—the actual execution layer where agents run. As agents proliferate, enterprises increasingly want to control AI at their own pace without being locked into any single vendor, and TrueForge aims to serve that need.
According to cost comparisons published by TrueFoundry, combining TrueForge with the open-source LLM GLM-5.2 achieves a 75% cost reduction compared to Anthropic's Claude Managed Agents (using Claude Opus 4.8). Specifically, when completing 11 out of 14 multi-step tasks spanning CRM, issue management, and document management systems in the enterprise benchmark 'DevRev's Enterprise-Bench,' the cost was 2.90 dollars with TrueForge versus 11.80 dollars with Claude Managed Agents. Additionally, when comparing with the same model (Claude Opus 4.8), TrueForge costs about 30% less at 8.50 dollars, according to the company.
Why offer it for free? Co-founder and COO Anuraag Gutgutia explained in a VentureBeat interview that 'many customers requested a mechanism for launching agents in managed form.' He also stated that 'this is not a replacement for existing harnesses like Claude, but rather an alternative option that runs in parallel and can be used in a vendor-agnostic, low-cost manner,' indicating the company anticipates coexistence with existing services.
In practical development scenarios, a key feature is the simplicity of running on a local environment with just a single command and SQLite. On the other hand, transitioning to production-like environments shared among teams is supported through a design that combines Docker Compose or Helm with Postgres and Redis. However, TrueFoundry explicitly states that local setup is intended only for individual developer machines and is not suitable for production services directly exposed to the internet.
TrueFoundry's business strategy involves using TrueForge as an entry point while driving adoption of the paid AI Gateway at the enterprise core. Gutgutia stated that 'regardless of which harness is used, we want all traffic to flow through our gateway,' positioning TrueForge as that gateway. As AI agents penetrate deeper into enterprise systems, the importance of cost management and access control increases. The release of TrueForge can be viewed as a strategic move in the broader infrastructure competition.
This article is an original work independently written and edited by the AI issue editorial team based on factual reporting. © AI issue. Unauthorized reproduction, redistribution, or use for AI training is prohibited.