Skip to content
AI.info

The Pulse

LangChain puts agent records at the center of LangSmith

LangSmith now groups an agent’s environments, traces and development tools under a single record. LangChain says the redesign also supports registration of agents running on other platforms.

LangChain puts agent records at the center of LangSmith

LangSmith now groups an AI agent’s environments, traces and development tools under a single record, replacing a project hierarchy that did not tie traces to a particular agent. LangChain detailed the redesign on November 16, saying the change connects work on building, running, improving and governing agents.

Each record has a named owner, a globally unique and immutable identifier, and a type that determines the dashboard information, data connections and evaluation criteria associated with the agent. Local, development, staging and production environments sit under the same agent ID.

From projects to agents

Previously, traces sent to LangSmith used a flat project hierarchy, and traces from different environments for the same agent were not associated with one another. Under the new model, environments and traces are grouped beneath an agent record.

An agent’s record provides access to its traces, evaluations, Agent Server and issues identified by LangSmith Engine. The company says the change is intended to bring together workflows that had been spread across separate tracing, monitoring and deployment surfaces.

Teams can describe an agent using a natural-language builder in LangSmith Studio, write it in code using LangChain’s open-source harnesses, or use another framework. Agents can run on LangSmith Agent Server or remain on an existing runtime.

Tracing, evaluations, datasets, annotation and Engine work with agents that send traces to LangSmith. LangChain says teams using both Engine and Agent Server can validate a proposed fix against an Agent Server preview before merging it into production.

Dashboards tailored by type

General-purpose and coding agent types are available, with a voice-agent type planned. Assigning a type supplies a preconfigured dashboard and evaluation criteria suited to that kind of agent.

The general-purpose dashboard includes an agent description, recent activity, online monitoring, commonly used models and tools, and failures surfaced by Engine. The coding-agent view tracks merged pull requests, token cost and the kinds of work completed.

Coding agents can also connect a GitHub repository to track outcomes of their use. LangChain says agents of the same type are measured consistently, allowing comparisons without first creating custom views.

Registration across runtimes

LangChain says it works for agents running on any platform that supports OpenTelemetry Protocol (OTLP), including labs and hyperscalers. Once connected, those agents can be registered alongside internally built agents.

Cost and quality metrics roll up to each agent across its environments. The agent registry and LLM Gateway provide organization-level visibility and strict controls over every registered agent.

The new LangSmith is available globally in cloud, bring-your-own-cloud and self-hosted environments.

What remains on the roadmap

Centralized dashboards are currently available at workspace level. LangChain is working on organization-level dashboards for aggregated performance metrics and trends across a company.

The company also plans to enable agent registration with an organization’s identity provider, so security teams can manage and audit agents as identities. LangChain says it is building toward a workflow in which reviewed production traces flow into evaluation datasets and trajectories from the best runs become fine-tuning data; it says each of these pieces exists in LangSmith today and it is now connecting them.

Sources

Companies in this story

Related stories

Explore

More articles