Agents

An agent is an AI system that completes tasks end to end using tools, skills, and subagents. In LangSmith, an agent is also the unit a workspace is organized around. Name it once, and LangSmith records what that agent does under that name, divided across four environments: Production, Staging, Development, and Local. A trace, an alert, an evaluator, and a prebuilt dashboard each belong to one environment of one agent. A dataset and a custom dashboard are the exceptions. A dataset attaches to one or more agents and to no environment, and a custom dashboard attaches to an agent rather than to an environment. Every trace carries its agent and one of that agent’s environments, so what produced a trace and where it ran are recorded as it arrives. LangSmith can then act on that context: an alert watches production and stays quiet about local runs, each environment gets its own prebuilt dashboard, and an Insights report covers the traffic of the environment you run it on. Every trace maps to the agent that produced it and the environment it ran in.

​

Environments

An environment records where an agent was running when it produced a trace. The names are fixed: Production, Staging, Development, and Local. For more information, see Agent environments.

​

Identifiers and display names

Every agent has two names, and they do different work:

• Identifier: Set when the agent is created, and not editable afterwards. Whether you supply it or LangSmith generates it depends on how the agent was created. LangSmith uses it to address traces, so it stays stable even when the display name changes.
• Display name: Shown wherever the agent appears in the UI. For an agent built in the UI, rename it from the Name field in its Build configuration. Renaming does not move existing traces or change where new traces land.

The identifier is also what the browser address bar carries, so you can read it from there or copy it from the ID field on the agent’s Overview.

​

Next steps

---