By admin 
In a rapidly evolving landscape where new AI agents emerge with daily frequency, San Francisco-based B2B machine learning startup TrueFoundry has addressed a growing enterprise challenge: providing developers with enhanced control over their burgeoning AI agent ecosystems while simultaneously driving down operational costs. The company, co-founded in 2021 by former Meta engineers, has officially released its custom AI agent harness, aptly named TrueForge, under the permissive MIT License. This strategic move makes the powerful framework available on GitHub, empowering developers and enterprises to integrate it seamlessly with their preferred AI models, fork it for customization, modify it to their specific needs, self-host it for maximum control, and even incorporate it into their commercial products without restrictive licensing.
TrueForge represents a significant stride towards democratizing advanced AI agent orchestration, offering a vendor-neutral and cost-effective alternative to proprietary solutions. The company’s commitment to transparency and developer enablement is underscored by this open-source release, signaling a departure from closed-ecosystem approaches.
The efficacy and cost-efficiency of TrueForge have been rigorously tested and validated by TrueFoundry itself. In a detailed blog post, the company reported on a benchmark test where TrueForge, when paired with the open-source GLM-5.2 Large Language Model (LLM), successfully completed 11 out of 14 complex tasks on DevRev’s Enterprise-Bench. This benchmark specifically evaluated multi-step tool utilization across critical enterprise systems, including Customer Relationship Management (CRM), issue tracking, and document management platforms. The results were striking: the entire process using TrueForge incurred a cost of just $2.90, a remarkable 75% reduction compared to the $11.80 expenditure required to achieve the same outcomes with Anthropic’s Claude Managed Agents harness, which leverages the powerful Claude Opus 4.8 model.
Further substantiating TrueForge’s cost advantages, TrueFoundry conducted a comparative analysis using the same underlying model, Opus 4.8, across both TrueForge and Claude Managed Agents. Even in this scenario, TrueForge demonstrated a significant cost saving of approximately 30%, with the execution costing $8.50 on TrueForge versus $11.80 on Claude Managed Agents. This consistent cost reduction highlights TrueFoundry’s focus on optimizing resource utilization and minimizing the financial burden associated with large-scale AI agent deployments.
The strategic decision to offer such a robust and cost-effective harness as open-source stems from direct customer demand. "We’ve had this ask from a bunch of customers," explained Anuraag Gutgutia, TrueFoundry’s co-founder and Chief Operating Officer, 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."
This statement underscores TrueFoundry’s strategic vision: TrueForge is positioned not as a direct competitor to existing managed services, but as a complementary solution that enhances flexibility and cost control within the broader AI agent ecosystem. TrueFoundry already offers a premium "AI Gateway" for enterprises seeking centralized control over model and Multi-Cloud Platform (MCP) access, credentials, permissions, budgets, and observability. TrueForge, on the other hand, operates at a higher level, managing the crucial "loop" where an AI model iteratively reasons, calls upon various tools, receives and processes the results, and continues its task execution until completion.
For enterprise developers, the practical implications of TrueForge are substantial. The harness facilitates a streamlined development workflow, allowing developers to begin locally with a single command and a lightweight SQLite database for rapid prototyping and testing. As their projects mature, they can effortlessly transition the same agent harness into a shared deployment environment using familiar tools like Docker Compose or Helm, integrating with robust databases such as PostgreSQL and Redis for scalable production use. TrueFoundry, however, prudently cautions that the local configuration is intended solely for individual developer use and should not be exposed as an internet-facing production service due to security considerations.
Gutgutia elaborated on TrueFoundry’s overarching ambition: "We’ve had this ask from a bunch of customers," he 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." The ultimate goal is for TrueFoundry’s AI Gateway to become the ubiquitous central layer orchestrating all AI traffic, irrespective of the specific agents and harnesses an enterprise chooses to employ. "There will be a set of companies that will use our harness as the way to launch managed agents," Gutgutia predicted, "while others may continue using Claude, other open-source harnesses or internal systems. But all that traffic should still be flowing through our gateway."
A key innovation within TrueForge lies in its sophisticated approach to context management, a critical factor in controlling the information flow back to AI models and thereby minimizing computational waste and cost. The architecture is meticulously designed to optimize the context window at every stage of an agent’s execution. This includes intelligently delaying the loading of MCP tool schemas until they are actively required, delegating isolated tasks to specialized sub-agents for more efficient processing, offloading excessively large tool results into files rather than overwhelming the active context, processing structured data outputs through dedicated code execution, and implementing automatic compaction mechanisms for lengthy conversations.
The default context compaction threshold is set at a substantial 50,000 tokens, a figure that can be fine-tuned on a per-agent basis to suit varying workload requirements. Furthermore, TrueForge adopts a distinct approach to sandboxing compared to traditional runtimes that maintain agents within isolated environments throughout their entire lifecycle. Instead, TrueForge keeps the core agent loop resident on the TrueFoundry server. A sandbox is provisioned dynamically as a tool only when the agent necessitates code execution or file manipulation. This innovative strategy significantly reduces unnecessary compute overhead, enabling a single server to manage a greater number of agents concurrently, further contributing to cost savings and improved resource utilization. TrueFoundry firmly asserts that these architectural choices directly translate into reduced model spending for enterprises.
The release of TrueForge positions TrueFoundry as a significant player in the AI orchestration space, offering a compelling alternative to established solutions. While proprietary managed agent harnesses like those offered by Anthropic provide a streamlined, end-to-end solution, they often come with a premium price tag and vendor lock-in. Open-source alternatives, while offering flexibility, may lack the robust governance, enterprise-grade features, and integrated cost management that TrueFoundry aims to provide. TrueForge appears to strike a balance, delivering a highly efficient and customizable runtime with the potential to integrate with TrueFoundry’s commercial AI Gateway for comprehensive enterprise control.
The distinction between using TrueForge independently and integrating it with TrueFoundry’s commercial AI Gateway is paramount for enterprise adoption. While the open-source harness offers unparalleled flexibility and can operate autonomously, it does not inherently provide the enterprise-grade access policies and governance controls that are crucial for large organizations. "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 advised. When paired with the AI Gateway, however, TrueFoundry claims that agents can seamlessly inherit existing enterprise identities and access controls applied to models, MCP servers, tools, skills, and other agents. Gutgutia elaborated that the gateway serves as the central nexus for enforcing enterprise Single Sign-On (SSO), identity providers, and granular permissions, eliminating the need for redundant reimplementation across every individual agent. This clear delineation between the free, open-source runtime and the commercial governance layer is a strategic move designed to appeal to platform engineering teams seeking both control and cost-effectiveness.
The development of TrueForge benefited from early engagement with key industry players. NetApp, for instance, served as a beta user, contributing valuable requirements during the development phase. NetApp’s IT organization has reportedly leveraged the technology for critical functions such as incident response and accelerated ticket triage, while also enabling self-service access to internal agents for their developers. Automattic has also been identified as an early adopter, further validating the harness’s capabilities in real-world enterprise scenarios.
TrueFoundry’s journey began in 2021 with a core mission to assist enterprises in deploying and efficiently operating machine-learning models, encompassing Kubernetes-based model serving, training, and infrastructure management. The company’s founders, Nikunj Bajaj, Abhishek Choudhary, and Anuraag Gutgutia, brought with them extensive experience from leading tech giants like Meta and financial institutions such as WorldQuant. Their shared insight was the critical need for mature systems and robust controls to prevent costly errors as AI systems moved from research labs into production environments within large corporations.
As generative AI rapidly permeated enterprise software stacks, TrueFoundry strategically expanded its MLOps foundation to encompass the management of LLM applications and, increasingly, the surrounding models, tools, and agents. By 2025, the AI Gateway had become a central pillar of TrueFoundry’s business model. This gateway acts as a sophisticated intermediary layer, connecting enterprise applications with model providers and handling essential functions such as request routing, authentication, access control, observability, budget management, and failover mechanisms.
This strategic evolution has been supported by substantial external investment. In February 2025, TrueFoundry secured a $19 million Series A funding round, led by Intel Capital, with significant 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 round brought the company’s total financing to approximately $21 million, as announced by Intel Capital. At the time of this announcement, TrueFoundry reported a fourfold year-over-year increase in its customer base and was managing over 1,000 clusters for machine-learning workloads.
The company’s growth trajectory has been further marked by an increasing focus on large-scale enterprise AI traffic. In January 2026, VentureBeat reported on TrueFoundry’s launch of TrueFailover, a solution designed for automatic rerouting of enterprise AI traffic. This coverage highlighted that TrueFoundry had surpassed 30 paid customers globally, achieved over $1.5 million in annual recurring revenue in the preceding year, and was processing more than 10 billion requests per month through its AI Gateway. Notable customers and deployments cited by TrueFoundry have included industry leaders such as NetApp, Siemens Healthineers, ResMed, Automation Anywhere, Nvidia, and Games24x7. In his interview with VentureBeat, Gutgutia further specified that NetApp, Siemens, Synopsys, and Automation Anywhere are among the Fortune 1000 organizations currently collaborating with TrueFoundry.
TrueFoundry has also strategically expanded its capabilities through acquisitions. In June 2026, the company acquired UK-based Seldon AI, a long-standing vendor in the MLOps space renowned for its Seldon Core software, widely used for production model serving and inference. This acquisition signifies TrueFoundry’s commitment to integrating traditional ML infrastructure with the rapidly advancing fields of LLMs, tools, and agents, aiming to provide a unified deployment and governance layer across these diverse AI components.
While TrueFoundry’s commercial focus has historically been on the control plane beneath enterprise AI workloads – managing access, routing, policy enforcement, cost monitoring, and service availability – TrueForge extends this strategy upward to the agent runtime itself. This open-source runtime provides a crucial execution layer where agents can actively perform their tasks. Gutgutia emphasized the complementary nature of this approach: enterprises can opt to run TrueForge independently or continue utilizing other agent harnesses, while TrueFoundry’s long-term business strategy centers on providing the overarching governance and infrastructure that underpins whichever agent solutions enterprises ultimately choose. This dual approach of offering both a powerful open-source tool and a comprehensive commercial control plane positions TrueFoundry as a versatile and strategic partner for enterprises navigating the complex and dynamic world of AI agent deployment and management.
