LangChain Inc., SF, founded 2022 by Harrison Chase, Series B 2025 (led IVP, with CapitalG/Sapphire/Sequoia/Benchmark, ~$125M). Agent engineering platform: tracing + evals + prompt hub + managed deployment for LangChain/LangGraph (framework-neutral SDK too). No acquisition, no sunset. Category ai-productivity-data, pipeline_stage=none. File under Dev (LLM-eval/observability sublane, sibling to Braintrust).Official site: https://www.langchain.com (LangSmith product) / direct: https://www.langchain.com/products/langsmithPricing: https://www.langchain.com/pricing

LangSmith — LangChain-Native Agent Engineering Platform for Tracing, Evals, Prompt Hub & Managed Deployment

Snapshot

Field	Value
Status	Active (changelog Aug 2026: Engine >2x issue detection, Tuned Evaluators, BYOC GA on AWS, Managed Deep Agents beta)
Origin	LangChain Inc., SF, 2022, Harrison Chase
Lane	LLM/agent observability + evaluation + prompt registry + managed agent runtime
Center of gravity	Trace-first (production monitoring), eval integrated, LangChain/LangGraph zero-config
Customers	~1,000 paying; 35% Fortune 500 incl. Workday, Cloudflare, ServiceNow, Klarna, Podium, Notion, Stripe-style
Compliance	SOC 2, HIPAA (Ent), GDPR, DPA, US/EU data residency, SSO/SAML/RBAC (Plus+), self-host/BYOC (Ent)
Category	`ai-productivity-data` / Dev (eval/obs sublane)

Features (table)

Layer	Capability	Notes
Tracing	Hierarchical traces (prompt/tool/retrieval/LLM/sub-agent), `@traceable` decorator, OTel-compatible, Py/TS/Go/Java SDK	Zero-config auto-instrument for LangChain/LangGraph via env var
Observability	Insights (trace clustering, failure-mode detection), cost tracking, latency/token breakdown, monitoring + alerting	Engine clusters traces → prioritized issues + suggested PR fix
Evals	Offline (dataset vs prompt/model versions) + online (live traffic scoring), LLM-as-judge / code / pairwise / human annotation queues	Statistical grounding on score deltas (vs Braintrust gap)
Prompt Hub	Versioned prompts, env labels, Playground A/B, AI-assisted prompt rewrite (Polly), Assistants runtime config	Runtime param injection without redeploy
Deployment	LangSmith Deployment: managed runtime for stateful LangGraph agents, durable exec, human-in-the-loop pause/resume, 1-click GH deploy, Fleet (no-code agents), Sandboxes (ephemeral code exec), expose agent as MCP server	Only platform in this lane with agent runtime
Engine	Autonomous trace monitoring → cluster behaviors → diagnose → recommend prompt/code fix → auto-build offline eval	LCU-metered ($1.50/LCU)
Gateway	LLM Gateway (Jul 2026): provider routing, cache, runtime controls	Not a general APM
No	No OSS community edition (SDKs open, platform closed), no non-LLM APM, no pure score-metering like Braintrust

Pricing (2026-09, USD)

Tier	Platform fee	Included	Overage / notes
Developer	$0	1 seat, 5,000 base traces/mo (14-day retention), community support	Hard-capped 5k until CC added; no extended retention; no deployment/Fleet/Engine
Plus	$39/seat/mo	Unlimited seats ($39 ea), 10,000 base traces/mo (14-day), 1 dev deployment (unlimited dev runs), Fleet 500/mo, Engine/Sandboxes/Insights, email support	Base trace overage $2.50/1k (14-day); extended 400-day $ 5.00/1k; deployment run $0.005; uptime$ 0.0036/min prod; Fleet run $0.05; Engine$ 1.50/LCU; Sandboxes vCPU/GiB hourly
Enterprise	Custom (annual)	Custom trace volume, SSO/SAML, SCIM, RBAC, audit, BYOC/self-host (AWS/GCP/Azure VPC), BAA, SLA, startup credits	—

starting_price=$0; first paid $39/seat/mo. Bill = seats ×$ 39 + trace meter + retention choice + optional Engine/Deployment/Fleet/Sandbox usage. LLM token cost billed by your model vendor, not LangSmith.

Reviews (2026)

Source	Signal
LangChain vs Braintrust page (vendor)	LangSmith wins: LangChain-native zero-config tracing, Engine auto-fix PRs, managed LangGraph deploy, annotation queues with approval gates, statistical eval grounding
genai.qa / aitoolsatlas / morphllm	Praised: deepest LangChain/LangGraph instrumentation, full trace→eval→deploy loop, Fleet/Sandboxes. Criticized: per-seat × trace-meter double bill shocks scaling teams, non-LangChain stacks lose auto-instrument polish, some eval features need extra LLM spend, closed-source (vs Langfuse)
TopReviewed panel	8.2/10 — “full trace-to-eval-to-deploy loop”, dinged on shallow attribution outside LangChain
Case studies	Klarna 80% faster resolution / 70% automation; Podium 98.6% F1 on agent eval

Praised: LangGraph deploy + Engine flywheel + annotation workflow + US/EU residency. Complained: $39/seat ×$ 2.50/1k trace compound cost, 5k free traces gone in 1–2 days in prod, lock-in perception, no OSS self-host.

Access / Integration (table)

Method	Supported
SDK	Python, TypeScript, Go, Java
Auto-instrument	LangChain, LangGraph (env var), OpenAI/Anthropic/Vercel AI SDK via wrappers, OTel
Deployment	LangSmith Deployment (LangGraph agents), Fleet (no-code), Sandboxes, MCP server expose
CI	Custom GitHub Actions (less native than Braintrust’s block-merge action)
Hosting	Cloud (US/EU), Hybrid (SaaS control + self-host data), Fully self-hosted (Ent)
No	No REST “score text” free endpoint, no frontend product analytics

Best For / Not For

Best For	Not For
LangChain/LangGraph production agents (zero-config tracing)	Framework-agnostic eval-first teams (→ Braintrust/Langfuse)
Teams wanting managed stateful agent deploy + tracing + evals in one	Tiny side-projects (5k trace cap dies in days)
Enterprises needing SSO/SAML/SCIM/BYOC/self-host + SLA	OSS-self-host-only mandates (→ Langfuse/Phoenix)
Human-in-the-loop agent workflows (pause/resume, annotation queues)	PM-only prompt shops unwilling to pay per-seat $39
Klarna/Podium-scale trace volume with regression gates	Non-LLM microservice APM (→ Datadog)

Competitors (lane map)

Lane	Tool	LangSmith edge / gap
Eval-first platform	Braintrust	Braintrust scores-meter + Loop + framework-agnostic; LangSmith trace-meter + Engine auto-fix + LangGraph deploy
OSS observability	Langfuse, Arize Phoenix	OSS self-host free; LangSmith managed + agent runtime + LangChain-native
Prompt mgmt	Humanloop, PromptLayer	Lighter; LangSmith deeper trace→eval→deploy
Gateway/obs	Portkey, Helicone	Routing focus; LangSmith eval/deploy broader
CI prompt test	Promptfoo	YAML red-team OSS; LangSmith production flywheel
APM legacy	Datadog LLM Obs, Fiddler	General APM; LangSmith agent-native + LangGraph runtime
In-list overlap	Exa, Cody, JB AI, Continue, Aider, Cline, Mintlify, Braintrust, Q Dev(sunset), Phind(shutdown)	Dev lane adjacent (eval/obs vs coding/docs)

Tags / Keywords

langsmith,active-2026-09,llm-observability,agent-tracing,evals,langchain-native,langgraph-deploy,prompt-hub,engine-auto-fix,annotation-queues,fleet,sandboxes,mcp-expose,soc2-hipaa-gdpr,developer-free,plus-39-seat,series-b-125m,harrison-chase,not-shutdown,ai-productivity-data,pipeline_stage=none,competitors-braintrust-langfuse-phoenix-humanloop,dev-eval-sublane

Want me to fold LangSmith (re-verified) into the locked slug/name/short_desc + structured fields + _event_log(none) batch alongside the eight active Dev/audio/video cards + Phind(shutdown) + Q Developer(sunset) + Braintrust?