LangSmith
by LangChain
The observability, evaluation, and deployment platform for LLM applications and agents — debug, test, and improve AI systems in production.
See https://smith.langchain.com
Features
- Tracing: structured run logs for LLM apps with typed renderers for LLM, retriever, tool, chain, prompt, and parser run types
- Filtering and debugging traces by run type, time, and custom metadata
- LLM Playground — bring any traced LLM call into an interactive sandbox for prompt iteration
- Multi-turn evaluations — automated and human evaluation of agent conversations
- LangSmith Agent Builder — no-code natural-language agent creation with memory and metaprompting
- Agent memory: agents store interactions and update their own instructions over time
- LangSmith Deployments — one-click hosting with automatic A2A (Agent-to-Agent) endpoints
- A2A protocol support: every deployed agent gets an A2A-compatible endpoint automatically
- “Poly” AI assistant — summarizes traces and answers questions about runs
- LangSmith Fetch — pull traces into the terminal for coding agents to inspect programmatically
- 7,500+ Arcade dev tools available in LangSmith Fleet
- Insights Agent — AI-generated analytics over your traces and evaluations
Superpowers
LangSmith turns opaque LLM application behavior into readable, filterable, actionable traces — the difference between raw logs and purpose-built debugging UX. The playground button on any traced LLM call lets you immediately iterate on prompts without rewriting code. Agent Builder takes it further: describe an agent in plain language, it identifies required tools, asks follow-up questions, and builds the prompt — then stores behavior changes in memory so the agent improves over time. LangSmith Deployments plus A2A protocol makes it trivial to wire deployed agents together into multi-agent systems. Best suited for teams building production LLM pipelines who need observability, evaluation automation, and iterative improvement loops.
Pricing
- Free developer tier (limited traces/month)
- Plus and Enterprise plans with higher trace volume, evaluation credits, and team features
- LangSmith Agent Builder in private beta (waitlist)
- Self-hosted deployment option available