Quick Facts
- TrueForge completed 11 of 14 tasks on DevRev’s Enterprise-Bench for $2.90 using GLM-5.2, compared to $11.80 for Claude Managed Agents running Claude Opus 4.8.
- The platform ships under an MIT license and supports OpenAI, Anthropic, Google Gemini, and more than 20 additional model providers out of the box.
- Enterprise customers NetApp and Automatiq are already running agentic workloads on TrueForge, with NetApp using it for incident response and ticket triage.
TrueFoundry launched TrueForge on Aug. 19, an open-source agent harness built for enterprise teams that want to build, deploy, debug, and govern production AI agents without binding themselves to a single model provider.
The company benchmarked TrueForge against Anthropic’s Claude Managed Agents on DevRev’s Enterprise-Bench, a test of multi-step tool use across CRM, issue tracking, and document management systems. Paired with the open-source GLM-5.2 model, TrueForge completed 11 of 14 tasks for $2.90. Claude Managed Agents running Opus 4.8 cost $11.80 for the same results — a 75% difference.
When the same model, Opus 4.8, ran inside each harness, TrueFoundry still claims a 30% cost advantage: $8.50 for TrueForge versus $11.80 for Claude Managed Agents.
What TrueForge Does
TrueForge runs the full agent execution loop — model calls, MCP tools, skills, sandboxing, human-approval workflows, context management, and session state. It exposes that loop three ways: a chat UI, an HTTP API with a TypeScript SDK, and an embeddable UI SDK.
The platform ships with 40-plus built-in tools and web search powered by Tavily. Every model call routes through TrueFoundry’s AI Gateway, which already processes more than 1 trillion tokens a day for enterprise customers, enabling budget enforcement, rate limits, and policy guardrails.
TrueFoundry released TrueForge under an MIT license on GitHub. The entire runtime is in the open-source repository. A hosted, pay-per-usage version is also available for teams that prefer not to manage their own infrastructure.
Why It Matters for Enterprise Buyers
Claude Managed Agents only runs Claude models on Anthropic’s cloud. TrueForge runs any model on the operator’s own infrastructure. That distinction is central to TrueFoundry’s pitch.
“We’ve had this ask from a bunch of customers,” said Anuraag Gutgutia, co-founder and COO of TrueFoundry. “It is not a replacement. People will use this alongside other harnesses, like the cloud-managed ones or the commercial-provider-managed ones, but this will serve as a way for people to use them in a vendor-neutral way and also at a lower cost.”
NetApp, a beta user of TrueForge, contributed requirements during development. Robert Rubin, senior director of platform engineering at NetApp, said TrueFoundry has become “the central platform in IT where agentic apps and agents are routed through” and described it as changing how quickly the team can move an agent from idea to production scale.
NetApp’s IT organization has used the technology for incident response and faster ticket triage, and has exposed internal agents as self-service tools for developers. Automatiq is also running agentic workloads on the platform.
The Broader Bet
Open models such as GLM-5.2 are closing the performance gap with frontier proprietary models at a fraction of the cost. Most managed agent platforms still lock enterprises into a single vendor’s models and pricing structure.
TrueFoundry co-founder and CEO Nikunj Bajaj, who previously built AI infrastructure at Meta, said teams outside big tech are still forced to choose between control and convenience the moment they hit production. TrueFoundry’s position is that the agent harness layer is now a strategic control point for enterprises, especially as AI agents move from developer laptops into customer-facing products and shared workflows.
TrueFoundry’s own internal agent, Ask TFY, which enterprise teams use to manage AI Gateway configurations and troubleshoot agent workflows, is built on the same TrueForge harness.
This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.