TrueFoundry launches TrueForge, open source rival to Claude agents

TrueFoundry released TrueForge today, an open-source framework for building and deploying AI agents that directly competes with Anthropic's Claude Managed Agents. The Bengaluru-based MLOps company is betting that enterprises want agent infrastructure they can run on their own terms, not locked into a single model provider.

What TrueForge actually does

TrueForge is an agent orchestration layer that sits between your application and the LLM of your choice. Unlike Claude Managed Agents, which ties you to Anthropic's models, TrueForge lets teams swap in Claude, GPT-4, Gemini, or open-weight models like Llama and Mistral without rewriting agent logic.

The framework handles the plumbing most teams struggle with: tool calling, memory persistence, multi-step reasoning traces, and guardrails. TrueFoundry designed it to run on any Kubernetes cluster, so teams already using their MLOps platform can deploy agents alongside existing model inference workloads.

This is the same playbook that worked for LangChain and LlamaIndex, but TrueFoundry is aiming higher in the stack. Where those frameworks focus on chains and retrieval, TrueForge targets production-grade agent deployment with built-in observability, versioning, and rollback.

Why open source matters here

Anthropic launched Claude Managed Agents earlier this year as a turnkey solution. You describe what you want the agent to do, and Claude handles execution. It works, but only with Claude models, and only on Anthropic's infrastructure. For regulated industries or teams with existing cloud contracts, that is a dealbreaker.

TrueForge ships under an Apache 2.0 license. Teams can inspect the code, modify agent behavior, and audit exactly how tools get called. That audit trail matters when an agent is making API calls that move money or access customer data.

The trade-off is obvious: you own the infrastructure burden. TrueFoundry's commercial offering handles that for teams who want managed hosting, but the core framework remains open.

The real competition

TrueFoundry is not the only company chasing this market. Microsoft's AutoGen, CrewAI, and dozens of smaller frameworks all promise multi-agent orchestration. What sets TrueForge apart is TrueFoundry's existing customer base in enterprise MLOps. They already handle model deployment, GPU allocation, and experiment tracking for companies running inference at scale.

Adding an agent layer on top of that infrastructure gives existing customers a natural upgrade path. For TrueFoundry, it is a land-and-expand play: teams using their platform for model serving can now build agents without adopting a separate stack.

What the announcement does not say

TrueFoundry's release focuses on capabilities, not benchmarks. There is no published comparison of agent latency, tool-call accuracy, or reasoning quality against Claude Managed Agents. That data will matter for teams deciding whether the flexibility of open source justifies leaving Anthropic's optimized stack.

Pricing for the managed version of TrueForge also remains unclear. TrueFoundry's existing platform charges based on compute and model inference, but agent-specific pricing, especially for features like memory persistence and long-running workflows, could change the calculus for budget-conscious teams.

The larger question is whether enterprises want to run agent infrastructure at all. Claude Managed Agents appeals precisely because teams do not have to think about scaling, memory stores, or failure modes. TrueForge is for teams who insist on control. That market exists, but it is smaller than the convenience market Anthropic is chasing.

Source: The New Stack / Adrian Bridgwater