TrueForge open-source AI agent harness challenges Claude Managed Agents with lower costs and vendor neutrality

TrueFoundry has released TrueForge, an open-source agent harness under the MIT license that lets enterprises run AI agents with any model they choose, self-hosted or in production. The company claims the harness completes the same multi-step tool tasks at 30% to 75% lower cost than Anthropic's Claude Managed Agents, based on internal benchmarks. For builders evaluating agent infrastructure, TrueForge offers a vendor-neutral alternative that pairs with TrueFoundry's commercial AI Gateway for centralized governance.

TrueForge: an open-source agent harness with a cost advantage

TrueForge is a general-purpose production agent harness designed to run the reasoning-tool loop that turns an LLM into a working agent. It is released under the MIT license, meaning developers can fork, modify, and incorporate it into commercial products without licensing fees. The harness supports local development with SQLite and production deployments via Docker Compose or Helm with Postgres and Redis VentureBeat.

TrueForge's architecture focuses on context engineering: it delays loading MCP tool schemas until needed, delegates isolated tasks to subagents, offloads large tool results to files instead of the context window, and automatically compacts long-running conversations at a default threshold of 50,000 tokens. The sandbox is provisioned only as a tool when the agent needs to execute code, rather than keeping the agent in an isolated environment throughout its run VentureBeat.

How the cost savings stack up

TrueFoundry published benchmark results on DevRev's Enterprise-Bench, which tests multi-step tool use across CRM, issue tracking, and document management systems. Using the open-source GLM-5.2 model, TrueForge completed 11 of 14 tasks at a cost of $2.90, compared to $11.80 for Claude Managed Agents using Claude Opus 4.8. When using the same model (Opus 4.8) in both harnesses, TrueForge still reported a cost of $8.50 versus $11.80 VentureBeat.

These figures are vendor-reported and have not been independently verified. Actual savings will vary by workload, model choice, and infrastructure costs.

Vendor-neutral design and enterprise governance

TrueForge is designed to work with any model provider or MCP server, making it a vendor-neutral alternative to managed runtimes like Claude Managed Agents. The harness itself is free, but TrueFoundry's commercial AI Gateway provides the enterprise governance layer: centralized SSO, identity-based permissions, budgets, observability, and failover routing The New Stack. Without the gateway, the open-source harness does not automatically enforce enterprise access policies. TrueFoundry co-founder Anuraag Gutgutia told VentureBeat that the gateway is where "enterprise SSO, identity providers and granular permissions can be centrally enforced rather than reimplemented separately for every agent" VentureBeat.

Early adopters include NetApp, which used TrueForge for incident response and ticket triage, and Automattic VentureBeat. TrueFoundry has raised about $21 million in funding and acquired Seldon AI in June 2026, signaling a strategy to unify ML, LLMs, tools, and agents under a common deployment and governance layer VentureBeat.

What to watch for

The cost and flexibility claims are compelling, but they come from company benchmarks and press coverage, not independent audits. The open-source harness is free, but the governance features that enterprises typically need require the commercial AI Gateway. Teams evaluating TrueForge should test it against their own workloads and compare total cost including model inference, infrastructure, and any gateway licensing. The harness is available on GitHub under the MIT license GitHub.