Running AI agents in production is getting expensive, fast. TrueFoundry thinks it has a fix: an open-source “agent harness” called TrueForge that sits between your AI models and the messy reality of enterprise deployment, promising to slash costs by as much as 75% compared to managed alternatives.
The company launched TrueForge on August 19, positioning it as a runtime layer that handles the plumbing most enterprises struggle with: model calls, tool management, session state, and the kind of centralized governance that keeps CISOs sleeping at night.
What TrueForge actually does
Instead of being locked into one provider’s ecosystem, TrueForge lets organizations swap between models and deployment environments without rewriting their stack.
The cost savings break down along predictable lines. Organizations using the same model they’d use on a managed platform like Anthropic’s Claude can expect around 30% savings, mostly from cutting out the managed-service markup. Switch to open models like GLM-5.2, and that number climbs to 75%.
TrueForge strips that away by letting teams run the same workloads on their own infrastructure, whether that’s on-prem servers, a virtual private cloud, a hybrid setup, or standard public cloud environments.
The harness integrates with TrueFoundry’s existing AI Gateway and MCP Gateway, which handle cost controls, observability, and governance across deployments. It also supports budget enforcement and sandboxed execution, two features that matter enormously when you’re running dozens of autonomous agents that each have the ability to rack up compute bills.
The vendor lock-in problem
TrueFoundry was founded in 2021 by Nikunj Bajaj, Abhishek Choudhary, and Anuraag Gutgutia. The company has raised over $21 million from investors including Naval Ravikant and Intel Capital, building out an enterprise AI infrastructure platform that claims to reduce cloud spend by 50% and increase GPU utilization by 80%.
TrueForge’s open-source approach, available on GitHub and the AWS Marketplace, supports Kubernetes-native deployments, speaking the language most enterprise DevOps teams already use.
The 30% to 75% cost reduction range is notable because it’s wide enough to be honest. The low end represents the floor for organizations that want to keep using premium proprietary models but ditch the managed service wrapper. The high end requires organizations to actually migrate workloads to open-source models, which involves its own engineering effort and potential quality tradeoffs.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
