By admin 
In a rapidly evolving AI landscape, where the proliferation of intelligent agents presents new challenges for enterprises, San Francisco-based B2B machine learning startup TrueFoundry has emerged with a significant offering: TrueForge. This custom-built agent harness, released under the permissive MIT License and available on GitHub, aims to address the growing need for greater developer control over AI agents and their associated tools, while simultaneously driving down operational costs. Founded in 2021 by a team of former Meta engineers, TrueFoundry’s latest innovation empowers developers to leverage their preferred AI models, whether open-source or proprietary, enabling them to fork, modify, self-host, and seamlessly integrate TrueForge into their commercial products.
The strategic release of TrueForge signifies a pivotal move by TrueFoundry, positioning itself as a facilitator of greater flexibility and cost-efficiency in the burgeoning agent-based AI ecosystem. According to a detailed blog post by the company, an independent evaluation using TrueForge, paired with the open-source GLM-5.2 LLM, demonstrated remarkable performance on DevRev’s Enterprise-Bench. This benchmark, designed to assess multi-step tool usage across critical enterprise systems like CRM, issue tracking, and document management, saw TrueForge successfully complete 11 out of 14 tasks. Crucially, the cost associated with these operations was a staggering 75% lower than achieving identical results using Anthropic’s Claude Managed Agents harness, powered by the sophisticated Claude Opus 4.8 model. The benchmark revealed a cost of $2.90 for TrueForge versus $11.80 for Claude Managed Agents, highlighting a substantial economic advantage.
Even when directly comparing the same high-end model, Opus 4.8, within both harnesses, TrueFoundry reports a significant cost saving of approximately 30% for TrueForge over Claude Managed Agents ($8.50 compared to $11.80). This consistent cost advantage, regardless of the underlying LLM, underscores TrueForge’s architectural efficiency in managing AI agent operations.
The decision to offer such a powerful tool as an open-source solution stems from direct customer demand, as explained by Anuraag Gutgutia, TrueFoundry’s co-founder and COO, in an exclusive interview. "We’ve had this ask from a bunch of customers," Gutgutia stated. "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." This sentiment emphasizes TrueFoundry’s commitment to providing an adaptable and cost-effective alternative within the existing AI infrastructure landscape.
TrueFoundry already offers a robust commercial solution, the AI Gateway, designed for enterprises seeking centralized control over model and MCP access, credentials, permissions, budgets, and observability. TrueForge, however, operates at a different layer of the AI stack. It focuses on the crucial "loop" that enables an AI model to repeatedly reason, invoke tools, process results, and iteratively work towards task completion. This distinction is vital: while the AI Gateway acts as the central nervous system for managing AI resources, TrueForge provides the engine that drives the intelligent execution of agents.
For enterprise developers, the practical implications of TrueForge are significant. The harness allows for an agile development workflow, starting locally with a single command and a lightweight SQLite database. As the agent’s functionality matures, it can be seamlessly scaled to shared deployments utilizing Docker Compose or Helm, integrating with more robust database solutions like Postgres and Redis. It is important to note that TrueFoundry explicitly advises against using the local configuration for internet-facing production services, emphasizing its intended use for development and testing environments.
Gutgutia articulated TrueFoundry’s overarching vision: to establish its AI Gateway as the universal layer beneath any agent and harness combination an enterprise chooses. "There will be a set of companies that will use our harness as the way to launch managed agents," he explained, acknowledging that others will continue to opt for solutions like Claude, various open-source harnesses, or bespoke internal systems. "But all that traffic should still be flowing through our gateway." This strategic approach positions TrueFoundry to capture value across a broad spectrum of enterprise AI deployments, regardless of the specific agent execution framework employed.
The core innovation behind TrueForge’s cost-saving capabilities lies in its sophisticated approach to context management, a critical factor in mitigating the expenses associated with large language models (LLMs). By meticulously controlling the amount of information fed back into the model at each step of an agent’s execution, TrueForge minimizes redundant processing and token usage. This intelligent context engineering encompasses several key strategies: delaying the loading of Machine Communication Protocol (MCP) tool schemas until they are actively required, delegating isolated tasks to specialized sub-agents, offloading large tool results into files rather than burdening the active context window, processing structured results programmatically, and implementing automatic compaction for lengthy conversations. The documentation specifies a default context compaction threshold of 50,000 tokens, a figure that can be customized per agent to fine-tune performance and cost.
Furthermore, TrueForge redefines the role of the sandbox environment. Unlike traditional runtimes that maintain an agent within an isolated sandbox for its entire operational lifecycle, TrueForge adopts a more dynamic approach. The primary agent loop resides on the TrueForge server, with a sandbox provisioned only when the agent needs to execute code or interact with files. This architectural choice significantly reduces unnecessary computational overhead, enabling a single server to manage a greater number of agents concurrently. TrueFoundry contends that these deliberate design choices directly translate into reduced model spending for enterprises.
The company has also provided a comparative analysis of TrueForge against other leading orchestration harnesses, highlighting its distinct advantages. While specific details of this comparison are presented in a table format (not directly reproducible in this text format but understood to cover aspects like Type/Focus, License, Price, Model Flexibility, Deployment, and Key Differentiator), the underlying message is clear: TrueForge offers a compelling blend of open-source flexibility, cost-effectiveness, and advanced features tailored for enterprise adoption.
A crucial distinction for enterprise buyers lies between the standalone open-source TrueForge harness and its integration with TrueFoundry’s commercial AI Gateway. While TrueForge can operate independently, it does not inherently inherit an organization’s enterprise-grade access policies. "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 clarified. This necessitates that organizations implementing the open-source version must implement their own governance and security measures.
However, when paired with TrueFoundry’s AI Gateway, agents can seamlessly inherit existing enterprise identities and access controls. This includes policies already established for models, MCP servers, tools, skills, and other agents. Gutgutia emphasized that the AI Gateway serves as the central enforcement point for enterprise Single Sign-On (SSO), identity providers, and granular permissions, eliminating the need to reimplement these critical security measures for each individual agent. This distinction is paramount for platform engineering teams evaluating the project, as it delineates the freely available software (TrueForge) from the commercial control plane (the AI Gateway) that provides enterprise-grade governance.
TrueFoundry has already garnered significant traction with early adopters. NetApp, a prominent beta user, actively contributed requirements during TrueForge’s development. NetApp’s IT organization has reportedly leveraged the technology for critical functions such as incident response and expedited ticket triage, while also making internal agents accessible as self-service tools for developers. Automattic has also been identified as an early adopter, underscoring the growing confidence in TrueFoundry’s solutions.
The company’s journey began in 2021 with a focus on helping enterprises deploy and manage machine learning models, encompassing Kubernetes-based model serving, training, and infrastructure management. The founders—Nikunj Bajaj, Abhishek Choudhary, and Anuraag Gutgutia—brought with them extensive experience from tech giants like Meta and WorldQuant. Their shared realization was the increasing importance of robust infrastructure and stringent controls as AI systems moved into production environments within large corporations, aiming to prevent costly errors. As generative AI gained momentum across enterprise software, TrueFoundry strategically expanded its MLOps foundation to encompass the management of LLM applications and the intricate ecosystem of models, tools, and agents surrounding them.
By 2025, TrueFoundry had firmly established its AI Gateway as the cornerstone of its business. This gateway acts as a crucial intermediary between enterprise applications and model providers, meticulously handling aspects like request routing, authentication, access control, observability, budget management, and failover mechanisms. The company’s growth trajectory has been significantly bolstered by external financing, securing approximately $21 million in funding. A notable $19 million Series A round in February 2025, led by Intel Capital, saw participation from existing investors Eniac Ventures and Peak XV’s Surge, alongside Jump Capital and prominent angel investors like Gokul Rajaram and Mohit Aron. This funding propelled the company’s total financing to around $21 million, as detailed in Intel Capital’s announcement. At the time of this funding, TrueFoundry reported a fourfold increase in its customer base year-over-year and was managing over 1,000 clusters for machine learning workloads.
The business model has increasingly pivoted towards managing large-scale enterprise AI traffic. Reflecting this growth, TrueFoundry’s January 2026 coverage in VentureBeat regarding the launch of TrueFailover highlighted the company’s expansion to over 30 paid customers globally, exceeding $1.5 million in annual recurring revenue in the prior year, and processing more than 10 billion requests monthly through its AI Gateway. Notable customers and deployments cited by TrueFoundry include industry leaders such as NetApp, Siemens Healthineers, ResMed, Automation Anywhere, Nvidia, and Games24x7. In his interview with VentureBeat, Gutgutia further specified NetApp, Siemens, Synopsys, and Automation Anywhere as among the Fortune 1000 organizations collaborating with the company.
TrueFoundry’s strategic expansion has also included key acquisitions. In June 2026, the company acquired UK-based Seldon AI, a seasoned MLOps vendor renowned for its Seldon Core software, which has been widely adopted for production model serving and inference. This acquisition signifies TrueFoundry’s intent to consolidate traditional ML, LLMs, tools, and agents under a unified deployment and governance framework, rather than treating them as disparate infrastructure categories.
The introduction of TrueForge represents a strategic extension of this vision, pushing the company’s influence upward into the agent runtime itself. Historically, TrueFoundry’s commercial focus has been primarily on the control plane situated beneath enterprise AI workloads—managing access to models and tools, routing requests, enforcing policies, monitoring expenditure, and ensuring service availability. TrueForge, however, provides an open-source runtime that complements this foundational layer, enabling agents to execute directly. Gutgutia articulated this relationship as synergistic: organizations can opt to run TrueForge independently or continue utilizing other agent harnesses, while TrueFoundry’s long-term business objective remains to offer the overarching governance and infrastructure that underpins any agent framework chosen by enterprises.
