LangChain launches Fleet for traced team agents

• LangChain has turned Agent Builder into Fleet, a LangSmith workspace for team-managed agents where the launch post says users can build agents in natural language, control who can edit or clone them, and route approvals through human-in-the-loop checks.
• The new platform centers on agent-scoped auth: in the product thread, LangChain describes “Agent Identity” as credentials and connections attached to the agent itself so teams can limit what it can read, write, and access across tools like GitHub and Slack.
• Fleet also ties execution back into LangSmith tracing and observability, with the announcement promising teams can “track and audit actions,” while the thread details position that as part of a broader push toward production monitoring for agents.

Fleet is a rebrand of Agent Builder, but the launch comes with a more opinionated enterprise packaging around multi-user agents. LangChain's launch post lists the core controls: natural-language agent creation, sharing permissions over who can “edit, run, or clone,” agent-specific authentication, human approval gates, and tracing through LangSmith Observability. The product video Fleet demo shows those controls inside a web workspace rather than as a loose collection of SDK features.

The companion feature thread adds details that matter for implementation. Fleet agents keep their own memory, can access “a collection of tools and skills,” and can be exposed through channels teams already use. LangChain also says the release includes credential management with “Claws” and “Assistants,” Google-Docs-style sharing controls, and custom Slack bots so each agent has its own identity in Slack. The product is live in the Fleet app, with a longer overview in the announcement post.

The clearest technical pitch is that Fleet wraps several hard production problems into one control plane. In the design thread, LangChain argues that agent deployments need an identity and security model “that reflect that” work is specified by humans but executed by agents, plus tooling for self-improvement, memory, evals, and external-system context engineering. That is a shift from single-user copilots toward long-lived agents with permissions, state, and measurable behavior over weeks or months.

LangChain is also framing observability as a distinct requirement rather than a logging add-on. The observability post says “you don't know what your agent will do until it's in production,” and that production monitoring for agents needs capabilities different from traditional software. Fleet's promise to audit actions inside LangSmith fits that framing: the product is less about spinning up one more bot and more about giving teams governed access, traceability, and channel-specific identities without hand-stitching auth and audit flows.