TrueFoundry has released TrueForge, an open-source, vendor-neutral AI agent harness licensed under the MIT License on GitHub. Designed to give enterprise developers greater control over agent tool orchestration while cutting costs, TrueForge achieves a 30% to 75% reduction in task completion expenses compared to Anthropic’s Claude Managed Agents.
Engineering Context Management to Eliminate Token Waste
The architectural core of TrueForge relies heavily on context engineering. Instead of dumping raw transcripts and expansive schema payloads into every single reasoning loop, TrueForge selectively controls what information returns to the language model. When managing multi-step workflows across systems like customer relationship management platforms, issue trackers, and document repositories, inefficient memory management quickly balloons API expenses.
To curb this waste, TrueForge delays the loading of Model Context Protocol (MCP) tool schemas until they are explicitly required by the workflow. It delegates isolated sub-tasks to smaller subagents, offloads oversized tool execution outputs directly into flat files rather than stuffing them into the active context window, processes structured results programmatically through native code, and automatically triggers compaction routines for long-running conversations. According to the platform documentation, the default compaction threshold is set at 50,000 tokens, though developers can reconfigure this limit on a per-agent basis.
Furthermore, TrueForge diverges from traditional execution runtimes that lock an agent inside a persistent sandbox throughout its entire lifecycle. The core orchestration loop remains resident on the primary TrueForge server, while a secure sandbox is provisioned strictly as an on-demand tool only when the agent needs to execute code or manipulate files. According to TrueFoundry, this separation minimizes unnecessary compute overhead and allows a single server instance to handle a significantly higher density of concurrent agent operations.
Benchmarking Costs Across Open and Proprietary Models
The economic impact of these architectural choices becomes clear when evaluating real-world multi-step tasks. In benchmark tests conducted on DevRev’s Enterprise-Bench—which measures agent capabilities across CRM, issue tracking, and document management systems—pairing TrueForge with the open-source GLM-5.2 LLM successfully completed 11 out of 14 tasks at a fraction of the cost required by proprietary alternatives.
Running the same workload through TrueForge powered by GLM-5.2 cost $2.90, representing a 75% savings compared to Anthropic’s Claude Managed Agents running on Claude Opus 4.8, which totaled $11.80. Even when utilizing the exact same underlying model—Claude Opus 4.8—inside the TrueForge harness, TrueFoundry reported a 30% reduction in expenses, bringing the cost down to $8.50 compared to the default Anthropic managed runtime.
“We’ve had this ask from a bunch of customers,” said Anuraag Gutgutia, TrueFoundry’s co-founder and COO, in an exclusive interview with VentureBeat. “You have an ability where you bring in agents and MCPs — can we also get something where you can actually launch these managed agents? I think that is the need we are satisfying. 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.”
Production Deployment Pathways and the Role of the AI Gateway
For development teams evaluating the project, the framework offers a clear migration path from local prototyping to scaled server deployments. Developers can initialize the harness locally using a single command backed by SQLite. However, TrueFoundry explicitly cautions that this local configuration is strictly intended for individual developer machines rather than internet-facing production environments.
For shared deployments, teams can transition the exact same harness into production using Docker Compose or Helm configurations backed by enterprise-grade data stores like PostgreSQL and Redis.
Despite being free and open source, TrueForge does not automatically enforce enterprise security policies out of the box. Gutgutia noted that organizations adopting the standalone open-source harness must implement access controls and governance rules independently or integrate them with internal authorization systems. To bridge this gap, TrueFoundry positions TrueForge to operate upstream from its commercial AI Gateway, a paid control plane that centrally manages model routing, credentials, user permissions, budgeting, guardrails, and observability.
“If you are using just the open source version of our agent harness, yes, you will need to put the right controls therein or in front of some other internal control system,” Gutgutia explained. When paired with the gateway, agents can automatically inherit enterprise single sign-on (SSO) identities and granular permissions tied to underlying models, MCP servers, and tools.
Enterprise Adoption and Market Positioning
Enterprise interest in the harness is already taking shape. NetApp participated as an early beta user during the development of TrueForge, utilizing the technology within its IT organization for automated incident response, rapid ticket triage, and exposing internal agents as self-service capabilities for software engineers. Automattic was also identified among the early adopters leveraging the platform.
Founded in 2021 by former Meta and WorldQuant engineers—Nikunj Bajaj, Abhishek Choudhary, and Anuraag Gutgutia—TrueFoundry originally established its footprint in Kubernetes-based machine learning operations (MLOps) infrastructure. The company expanded its focus toward generative AI orchestration, culminating in a $19 million Series A funding round in February 2025 led by Intel Capital, bringing its total financing to approximately $21 million. Building on that momentum, the firm acquired UK-based MLOps vendor Seldon AI to bolster its production model serving capabilities.
By coupling an MIT-licensed, vendor-neutral execution runtime like TrueForge with a centralized commercial gateway, TrueFoundry is betting that platform engineering teams will embrace a modular approach to artificial intelligence—allowing developers to swap out LLMs and agent harnesses freely while maintaining strict organizational oversight and predictable operating margins.
