TrueFoundry's open source AI agent harness TrueForge boasts 30%-75% cheaper task completion than Claude…

What happened

TrueFoundry's open source AI agent harness TrueForge boasts 30%–75% cheaper task completion than Claude Managed Agents

TrueFoundry, a San Francisco-based B2B machine learning startup co-founded in 2021 by former Meta and Google engineers, has released TrueForge, an open source AI agent harness designed to give enterprise developers tighter control over autonomous agents and tools while driving down the cost of task completion by 30% to 75% compared to Claude Managed Agents.

This article breaks down exactly what shipped, what changed for builders who are orchestrating AI agents in production, how to get started, what compatibility issues to watch for, and what developments are worth tracking next. It is aimed at platform engineers, ML infrastructure teams, and developers who are already running or evaluating agentic workloads and need cost-aware, controllable alternatives to managed offerings.

What shipped

How it works

TrueFoundry released TrueForge, an open source agent harness, as a standalone framework built to address the dual pressure of rising inference costs and insufficient developer control in enterprise agentic deployments. The project comes out of TrueFoundry's own work as a B2B ML platform startup founded in 2021, drawing on engineering experience from Meta and Google. TrueForge is positioned as an alternative to managed agent services, with the company's own benchmarking showing task completion costs that run 30% to 75% lower than Claude Managed Agents on representative workloads.

The harness gives teams the ability to define and manage both agents and the tools those agents can invoke, rather than delegating those decisions to an opaque hosted layer. The open source release reflects TrueFoundry's broader thesis that as agent counts scale inside an organization, the lack of observability and cost predictability in managed offerings becomes a structural problem rather than a temporary inconvenience.

What changed for builders

Builders who previously relied on Claude Managed Agents or similar hosted orchestration layers now have an open source path that keeps agent logic, tool definitions, and execution routing within their own infrastructure. TrueForge is designed to surface the decisions that managed services typically absorb — tool selection, retry behavior, agent handoffs — so that engineering teams can instrument, audit, and modify those paths directly.

Why it matters

The cost reduction of 30% to 75% comes from giving builders control over which model calls are made and when, rather than paying for an intermediate managed layer's overhead. For teams that are running many concurrent agents or processing high task volumes, this kind of cost structure is materially different from a per-use managed service. The tradeoff is that builders take on the responsibility of maintaining and configuring the harness rather than relying on a vendor to handle it.

How to install or upgrade

TrueForge is available as an open source project, meaning the starting point is the TrueFoundry repository and its accompanying engineering blog post at truefoundry.com/blog/engineering/trueforge. Installation follows the standard open source path: clone the repository, review the setup documentation, and configure your agent definitions and tool registries according to the project's interface conventions.

Who is affected

Teams that are currently running TrueFoundry's existing ML platform should check whether TrueForge is designed to slot into their existing infrastructure as a native component or whether it operates as a separate layer. Because TrueFoundry is a B2B platform company, enterprise customers may also have access to support channels or onboarding guidance beyond what is documented in the public repository. No specific version number has been disclosed for this initial release.

Gotchas and compatibility

The 30% to 75% cost improvement figure comes from TrueFoundry's own benchmarks against Claude Managed Agents, which means the actual savings a team sees will depend heavily on their specific workload, task complexity, and how efficiently they configure their tool and agent definitions. Builders should run their own representative tasks through both systems before treating these numbers as a guarantee.

Because TrueForge is open source and self-hosted, teams bear responsibility for security, scaling, and reliability in ways that managed agent services typically handle for them. If your organization has strict data-residency requirements or already maintains significant ML infrastructure, this is likely an advantage. If your team is small or infrastructure-light, the operational overhead of running your own agent harness is a real cost that does not show up in the per-task inference numbers.

What to watch next

What to watch next

TrueFoundry was co-founded by engineers who came out of Meta and Google, so the project's design decisions will likely reflect patterns from large-scale internal ML infrastructure at those organizations. Watching how the open source community responds to TrueForge's tool definition interface and agent orchestration model will indicate whether those patterns translate well to enterprise teams outside those environments.

The broader question is whether open source agent harnesses as a category can reliably undercut managed services on cost while remaining maintainable as model APIs evolve. TrueFoundry's release joins a growing set of frameworks making exactly this bet. Watch for benchmark reproductions from independent teams, community contributions to the repository, and whether TrueFoundry releases updated cost comparisons as Claude or other managed agent offerings change their pricing or capabilities.