Deep Agents/Protocols & Platform
Intermediate14 min

LangSmith Fleet: No-Code Agent Platform

Fleet is LangSmith's no-code agent builder (rebranded from Agent Builder, March 2026). The critical first decision is the authorization model: Assistants (user's OAuth, private inbox) vs Claws (fixed credentials, shared access). This article covers that decision, channels, schedules, Skills, the Tool Server, observability, permissions, cost model, and self-hosted deployment.

Quick Reference

  • Fleet = no-code agent creation within LangSmith (formerly Agent Builder, rebranded March 2026)
  • Two authorization types: Assistants (user's OAuth, per-user inbox) vs Claws (fixed credentials, shared access)
  • Channels: event-driven triggers from Gmail, Slack (dedicated bot handles), Teams, webhooks
  • Schedules: thread-bound (state accumulates across runs) vs stateless (fresh each run) cron tasks
  • Skills: shareable instruction sets with domain knowledge — AI-assisted, template, or manual creation
  • Tool Server: standalone MCP framework with Agent Auth; acts as MCP gateway for external services
  • Agent Inbox: centralized HITL — review, approve, or reject agent actions before they execute
  • Pricing (Apr 2026): Developer = 1 agent, 50 runs free; Plus = $39/seat, runs billed as LangSmith traces

Should You Use Fleet or Write Code?

Fleet is not a replacement for code-based agents — it covers a specific slice of the problem space. If the reader of this article needs to decide between Fleet and `create_agent` / `create_deep_agent`, the right question is not 'which is better?' but 'what does my use case actually require?'

Fleet UINo-code agent builder (LangSmith)Agent ServerAPI layer + orchestration runtimeLangGraphStateful agent runtimeChannelsGmail · Slack · TeamsWebhookSchedulesCron · DailyThread-boundSkillsInstruction setsDomain knowledgeTool ServerMCP frameworkAgent Auth + OAuthLangSmithtracesAgentInboxHITLFleet Webhooks → CI/CD publishing

Fleet stack — UI sits on Agent Server + LangGraph; Channels, Schedules, Skills, and Tool Server are the extension points

CriterionFleet (no-code)Code (create_agent / create_deep_agent)
Version controlConfig lives in UI only — no GitFull Git history, PR reviews, CI/CD
Custom state managementNot supportedFull LangGraph state, reducers, checkpointers
Multi-agent orchestrationNot supportedcreate_deep_agent + subagent spawning
Non-technical team membersSelf-service via Fleet UIRequires developer involvement
Event-driven automationChannels: Gmail, Slack, Teams, webhooksRequires custom webhook infrastructure
Time to first deployMinutes from templateHours to days for production code
Middleware / guardrailsNot customizableFull hook system, custom validators
Ops automation, quick prototypesExcellent fitOverkill for simple flows
Fleet agents cannot be version-controlled

Fleet configurations live in the LangSmith UI, not in a file. If your team requires Git-based review and rollback for agent changes, Fleet is the wrong choice — use code-based agents with the LangGraph CLI and deploy via LangGraph Platform.