LangChain uses Baseten Loops to train custom models for LangSmith Engine, the in-platform agent that helps users debug and improve their own AI agents autonomously. Baseten said on September 16, 2026 that the collaboration combines LangChain's tools for building, evaluating, and deploying AI agents at scale with Baseten infrastructure for model training and inference.

LangChain trains custom models for agent-specific tasks by fine-tuning large open-weight models on agent traces, which are records of how an agent handled a task. The article gives LangSmith Engine examples such as reading a connected GitHub repository to diagnose why an issue is happening, then drafting the prompt or code change that goes into a pull request. For narrower tasks, including categorizing traces by failure mode and severity or mapping traces to existing open issues, LangChain can switch models and train a smaller open-weight model such as Qwen.

Baseten Loops provides managed infrastructure for fine-tuning models through an API, with support for supervised fine-tuning, reinforcement learning, and long-context workloads. The system connects training to Baseten's inference platform, so checkpoints can be evaluated during training and deployed directly. A LangChain quote in the article says Loops gives the company control and iteration speed while training, and that using training and inference on the same platform gives a clear path from model development to production.

LangChain uses Baseten Loops to train custom models for LangSmith Engine, the in-platform agent that helps users debug and improve their own AI agents autonomously. Baseten said on September 16, 2026 that the collaboration combines LangChain's tools for building, evaluating, and deploying AI agents at scale with Baseten infrastructure for model training and inference.

LangChain trains custom models for agent-specific tasks by fine-tuning large open-weight models on agent traces, which are records of how an agent handled a task. The article gives LangSmith Engine examples such as reading a connected GitHub repository to diagnose why an issue is happening, then drafting the prompt or code change that goes into a pull request. For narrower tasks, including categorizing traces by failure mode and severity or mapping traces to existing open issues, LangChain can switch models and train a smaller open-weight model such as Qwen.

Baseten Loops provides managed infrastructure for fine-tuning models through an API, with support for supervised fine-tuning, reinforcement learning, and long-context workloads. The system connects training to Baseten's inference platform, so checkpoints can be evaluated during training and deployed directly. A LangChain quote in the article says Loops gives the company control and iteration speed while training, and that using training and inference on the same platform gives a clear path from model development to production.