One other day, one other new AI agent harness is launched.
Solely this time, it is one which goals to remedy a rising enterprise drawback as AI brokers proliferate: enabling larger developer management of brokers and instruments, whereas decreasing price.
TrueFoundry, a San Francisco B2B machine studying startup co-founded in 2021 by former Meta engineers, has launched its personal customized TrueForge harness underneath the permissive MIT License on Github. Thus, it may be used with any of a developer (or their dad or mum enterprise’s) most popular AI fashions, forked, modified, self-hosted and included into business merchandise.
The company states in a blog post that when it used TrueForge paired with the open supply GLM-5.2 LLM to efficiently full 11 of 14 duties on DevRev’s Enterprise-Bench — testing multi-step instrument use throughout CRM, situation monitoring, and doc administration methods — it price 75% lower than attaining the similar outcomes with Anthropic’s Claude Managed Brokers harness powered by Claude Opus 4.8 ($2.90 in contrast to $11.80).
Utilizing the similar mannequin in every harness, Opus 4.8, TrueFoundry nonetheless claims a value financial savings of roughly 30% utilizing TrueForge in contrast to Claude Managed Brokers ($8.50 vs $11.80).
Why is TrueFoundry giving this powerfully environment friendly harness away free of charge?
“We’ve had this ask from a bunch of consumers,” mentioned Anuraag Gutgutia, TrueFoundry’s co-founder and COO, in an unique interview with VentureBeat. “You’ve gotten a capability the place you herald brokers and MCPs — can we additionally get one thing the place you may truly launch these managed brokers? I believe that is the want we are satisfying. It is not a alternative. Folks will use this alongside different harnesses, like the cloud-managed ones or the commercial-provider-managed ones, however it will function a manner for folks to use them in a vendor-neutral manner and in addition at a decrease price.”
Certainly, TrueFoundry already affords a paid “AI Gateway” for enterprises centrally controlling mannequin and MCP entry, credentials, permissions, budgets and observability. TrueForge, against this, handles what occurs above that gateway: the loop that lets a mannequin repeatedly motive, name instruments, obtain outcomes and proceed working till a job is full.
For enterprise builders, the sensible proposition is that they’ll begin domestically with a single command and SQLite, then transfer the similar agent harness right into a shared deployment utilizing Docker Compose or Helm with Postgres and Redis.
TrueFoundry explicitly warns that the native configuration is meant just for use on a developer’s machine, not as an internet-facing manufacturing service.
Gutgutia mentioned the firm in the end desires its AI Gateway to turn into the frequent layer beneath whichever brokers and harnesses an enterprise chooses.
“There can be a set of firms that can use our harness as the manner to launch managed brokers,” he mentioned, whereas others might proceed utilizing Claude, different open-source harnesses or inner methods. “However all that site visitors ought to nonetheless be flowing by means of our gateway.”
Context administration is the place TrueForge tries to minimize waste
TrueForge’s structure facilities on context engineering — controlling how a lot information will get despatched again into the mannequin on each step of an agent run.
That features delaying the loading of MCP instrument schemas till they are wanted, delegating remoted duties to subagents, shifting outsized instrument outcomes into information as a substitute of stuffing them into the lively context window, processing structured outcomes by means of code, and mechanically compacting long-running conversations.
The documentation units the default compaction threshold at 50,000 tokens, although it may be modified per agent.
TrueForge additionally treats the sandbox in a different way from runtimes that hold an agent inside an remoted setting all through its run. The core agent loop stays on the TrueForge server; a sandbox is provisioned as a instrument solely when the agent wants to execute code or work with information. TrueFoundry says that reduces pointless compute and permits a server to run extra brokers concurrently.
The corporate argues these decisions straight scale back mannequin spending.
How TrueForge compares to Claude Managed Brokers and different main orchestration harnesses
Sort / focus
-
TrueFoundry TrueForge: Normal-purpose manufacturing agent harness designed for enterprise deployments.
-
DeepSeek Harness: Open-source agent harness, at present positioned as a developer preview.
-
OpenAI Codex CLI: Coding-focused agent harness designed primarily for software-engineering workflows.
-
LangChain Deep Brokers: Normal-purpose agent harness constructed on LangGraph.
-
Anthropic Claude Managed Brokers: Totally managed manufacturing agent runtime operated by Anthropic.
License
-
TrueFoundry TrueForge: MIT.
-
DeepSeek Harness: MIT.
-
OpenAI Codex CLI: Apache 2.0.
-
LangChain Deep Brokers: MIT.
-
Anthropic Claude Managed Brokers: Proprietary.
Worth
-
TrueFoundry TrueForge: The open-source harness itself is free. Mannequin, sandbox and infrastructure prices are separate. TrueFoundry additionally affords an optionally available business governance layer by means of its broader platform.
-
DeepSeek Harness: No harness license payment. Customers individually pay for no matter mannequin suppliers and infrastructure they use.
-
OpenAI Codex CLI: The CLI is open supply. Underlying mannequin/API or subscription prices are separate, OpenAI says round $100–$200 per developer monthly, though precise spending varies considerably with mannequin alternative
-
LangChain Deep Brokers: Open supply, with mannequin and infrastructure bills separate. LangChain additionally affords optionally available business companies by means of LangSmith.
-
Anthropic Claude Managed Brokers: Claude tokens consumed plus $0.08 per operating session-hour, with runtime metered to the millisecond.
Mannequin flexibility
-
TrueFoundry TrueForge: Vendor-neutral and designed round bring-your-own-model help.
-
DeepSeek Harness: Multi-provider and not restricted to DeepSeek fashions.
-
OpenAI Codex CLI: Helps configurable inference endpoints, together with OpenAI-compatible companies and local-model choices.
-
LangChain Deep Brokers: Broad multi-provider help by means of the LangChain ecosystem.
-
Anthropic Claude Managed Brokers: Claude-centric.
Deployment
-
TrueFoundry TrueForge: Can run domestically as a single course of with SQLite, then transfer right into a manufacturing deployment utilizing Docker Compose or Helm with Postgres and Redis.
-
DeepSeek Harness: Designed for native or self-hosted operation.
-
OpenAI Codex CLI: Primarily a neighborhood CLI expertise, alongside OpenAI-hosted Codex services and products.
-
LangChain Deep Brokers: May be self-hosted or deployed by means of LangChain and LangSmith infrastructure.
-
Anthropic Claude Managed Brokers: Anthropic manages the runtime and infrastructure.
Key options
-
TrueFoundry TrueForge: MCP and power orchestration, subagents, human approval checkpoints, persistent classes, context compaction, large-result offloading, Code Mode, generative UI, tracing and a sandbox-as-a-tool structure.
-
DeepSeek Harness: Pluggable fashions, instruments, session storage and agent loops, together with sandboxing, permissions, approval gates and abilities.
-
OpenAI Codex CLI: Agent loop, repository and file operations, shell execution, MCP instruments, sandboxing, permissions, approvals and context administration.
-
LangChain Deep Brokers: Planning, subagents, abilities, filesystem-based context administration, persistent reminiscence, human-in-the-loop controls, MCP help and a number of sandbox backends.
-
Anthropic Claude Managed Brokers: Managed execution environments, persistence, instruments, sandboxing and infrastructure for long-running brokers.
Key differentiator
-
TrueFoundry TrueForge: Its strongest distinction is the mixture of an open-source, vendor-neutral harness with a transparent path from native improvement to a shared manufacturing runtime, plus an optionally available enterprise governance airplane by means of TrueFoundry.
-
DeepSeek Harness: Emphasizes deep modularity. Main components of the runtime, together with fashions, instruments, storage and the agent loop, are designed to be replaceable plugins.
-
OpenAI Codex CLI: Stands out as a extremely developed software-engineering-specific harness fairly than a general-purpose enterprise agent server.
-
LangChain Deep Brokers: Advantages from the broader LangChain and LangGraph ecosystem and affords a mature open-source path for constructing general-purpose brokers.
-
Anthropic Claude Managed Brokers: Minimizes operational burden by having Anthropic handle the runtime, however trades that comfort for tighter mannequin and platform coupling.
Open supply does not mechanically imply ruled
For enterprise patrons, one in every of the most necessary distinctions is between TrueForge by itself and TrueForge linked to TrueFoundry’s business AI Gateway.
The open-source harness can run independently. But it surely does not magically inherit a company’s enterprise entry insurance policies on its personal.
“If you happen to are utilizing simply the open supply model of our agent harness, sure, you have to to put the proper controls therein or in entrance of another inner management system,” Gutgutia instructed VentureBeat.
When paired with TrueFoundry’s gateway, the firm says brokers can inherit the identities and entry controls already hooked up to fashions, MCP servers, instruments, abilities and different brokers. Gutgutia described the gateway as the place the place enterprise SSO, identification suppliers and granular permissions may be centrally enforced fairly than reimplemented individually for each agent.
That distinction is doubtless to be necessary for platform engineering groups evaluating the mission. TrueForge is free software program; TrueFoundry’s governance layer is the business management airplane round it.
TrueFoundry says NetApp was a beta consumer of the harness and contributed necessities throughout improvement. Gutgutia mentioned NetApp’s IT group has used the expertise for incident response and quicker ticket triage, whereas additionally exposing inner brokers as self-service instruments for builders. He additionally recognized Automattic as an early consumer.
Background on TrueFoundry and its enterprise to date
TrueFoundry was based in 2021 to assist enterprises deploy and function machine-learning fashions, together with Kubernetes-based mannequin serving, coaching and infrastructure administration.
Its three co-founders — Nikunj Bajaj, Abhishek Choudhary and Anuraag Gutgutia — beforehand labored at Meta and WorldQuant, respectively.
Gutgutia mentioned the founders’ frequent expertise was working round mature methods the place infrastructure and controls have been designed to forestall pricey errors — an thought they believed would turn into more and more necessary as AI moved into manufacturing inside massive firms.
As generative AI unfold by means of enterprise software program, TrueFoundry expanded from that MLOps basis towards managing LLM purposes and, more and more, the fashions, instruments and brokers round them.
By 2025, the firm had made its AI Gateway a central a part of the enterprise: a layer sitting between enterprise purposes and mannequin suppliers that handles routing, authentication, entry controls, observability, budgets, guardrails and failover.
That evolution has been backed by roughly $21 million in outdoors financing. TrueFoundry raised a $19 million Collection A in February 2025 led by Intel Capital, with participation from present traders Eniac Ventures and Peak XV’s Surge, in addition to Leap Capital and angel traders together with Gokul Rajaram and Mohit Aron. The spherical introduced complete financing to about $21 million, in accordance to Intel Capital’s announcement.
At the time, TrueFoundry mentioned its buyer base had grown fourfold 12 months over 12 months and that it was managing greater than 1,000 clusters for machine-learning workloads.
The enterprise has since turn into more and more oriented round large-scale enterprise AI site visitors. In VentureBeat’s January 2026 coverage of TrueFoundry’s TrueFailover launch, the firm mentioned it had greater than 30 paid prospects worldwide, had exceeded $1.5 million in annual recurring income throughout the prior 12 months and was processing greater than 10 billion requests monthly by means of its AI Gateway.
Clients and deployments cited by TrueFoundry have included NetApp, Siemens Healthineers, ResMed, Automation Wherever, Nvidia, Games24x7 and others; Gutgutia additionally named NetApp, Siemens, Synopsys and Automation Wherever amongst Fortune 1000 organizations working with the firm in his interview with VentureBeat.
TrueFoundry has additionally been increasing by means of acquisition. In June 2026 it acquired UK-based Seldon AI, a longtime MLOps vendor whose Seldon Core software program has been used for manufacturing mannequin serving and inference.
As the acquisition reveals, fairly than treating conventional ML, LLMs, instruments and brokers as separate infrastructure classes, TrueFoundry is attempting to put them behind a typical deployment and governance layer.
TrueForge extends that technique upward into the agent runtime itself. Till now, TrueFoundry’s business heart of gravity has largely been the management airplane beneath enterprise AI workloads — deciding which customers and purposes can entry which fashions and instruments, routing requests, imposing coverage, monitoring spend and holding companies out there.
TrueForge provides the firm an open-source runtime above that layer the place brokers can truly execute. Gutgutia described the relationship as complementary: organizations can run TrueForge independently or proceed utilizing different agent harnesses, whereas TrueFoundry’s longer-term enterprise alternative is to present the frequent governance and infrastructure beneath whichever brokers enterprises select.
Disclaimer: This article is sourced from external platforms. OverBeta has not independently verified the information. Readers are advised to verify details before relying on them.
