LangChain
Also known as: LangGraph, LangSmith
Open-source framework for building LLM-powered agents and multi-agent pipelines, with observability via LangSmith and graph orchestration via LangGraph.
LangChain is one of the most widely adopted open-source frameworks for building applications and agents powered by large language models. Created by the company of the same name and available in both Python and JavaScript under an MIT license, it gives developers standard building blocks for working with models, tools, memory, and control flow, so they can assemble agents without locking into a single model provider. In October 2025 the project reached a stable 1.0 release, marking a commitment to a steady API with no breaking changes until a future 2.0.
The 1.0 version refocused LangChain around the core agent loop. Its main entry point, create_agent, sets up a proven tool-calling agent in a few lines, and a new middleware system lets developers customize behavior, adding human approval steps, compressing long conversations, or stripping sensitive data, through composable hooks rather than rewriting core logic. With access to a large ecosystem of integrations, teams can swap models, tools, and databases without rewriting their application.
LangChain agents run on LangGraph, the company's lower-level orchestration framework and runtime. Where LangChain favors speed and simplicity, LangGraph gives fine-grained control through a graph-based execution model of state, nodes, and conditional routing, suited to complex, long-running, and multi-agent workflows. It brings durable state that persists automatically so an interrupted workflow resumes where it stopped, built-in persistence for multi-day or background processes, first-class human-in-the-loop pauses for review and approval, token-by-token streaming, and checkpointing that lets developers replay and rewind execution. Because LangChain is built on LangGraph, teams can start with high-level APIs and drop down to the graph layer as their needs grow.
Rounding out the stack is LangSmith, LangChain's companion platform for tracing, debugging, and evaluating agents, which makes evaluation a first-class part of development rather than an afterthought. Together these pieces are used to run production agents at companies including Uber, LinkedIn, and Klarna, and the framework is widely treated as a default foundation for building reliable, model-agnostic agentic systems.
Vendor details
Canonical URL
https://www.langchain.com/
Category
Agent infrastructure
Company status
independent
Use cases & customers
Target customers
Deployment options
In practice
You want to stand up a tool-calling agent fast without being married to one model provider. LangChain's create_agent gives you a working agent in a few lines, and you can swap models, tools, or databases without rewriting the app.
Your agent handles a multi-day approval process and can't lose its place if the server restarts. LangGraph's durable state and built-in persistence resume the workflow exactly where it stopped, with human-in-the-loop pauses for review.
You need to add a human approval step or strip sensitive data before a model call without touching core logic. LangChain's middleware hooks let you wrap that behavior around the agent as composable steps.
Agentic Index coverage score
11.5 / 14 capabilities · 82%
| Integrations & Tool CallingAgentic AI Index Expansion Evidence Report | Full |
|---|---|
| Workflow OrchestrationAgentic AI Index Expansion Evidence Report | Full |
| Knowledge Grounding & RAGAgentic AI Index Expansion Evidence Report | Full |
| Human Oversight & GuardrailsAgentic AI Index Expansion Evidence Report | Full |
| Security, Identity & GovernanceAgentic AI Index Expansion Evidence Report | Partial |
| Observability & AuditabilityAgentic AI Index Expansion Evidence Report | Full |
| Memory & State PersistenceAgentic AI Index Expansion Evidence Report | Full |
| Deployment & Data ResidencyAgentic AI Index Expansion Evidence Report | Full |
| Prebuilt Agents, Templates & PacksAgentic AI Index Expansion Evidence Report | Partial |
| Triggers & Channel CoverageAgentic AI Index Expansion Evidence Report | Partial |
| Model Flexibility & RoutingAgentic AI Index Expansion Evidence Report | Full |
| APIs, SDKs & MCP ExtensibilityAgentic AI Index Expansion Evidence Report | Full |
| Testing, Debugging & OptimizationAgentic AI Index Expansion Evidence Report | Full |
| Browser & Computer UseAgentic AI Index Expansion Evidence Report | Unable to verify |
The Agentic Index coverage score grades every vendor Full, Partial or Unable to verify against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
LangChain released Deep Agents v0.7, an update to its open-source agent harness that reduces base input tokens by approximately 65 percent. The release removes the default system prompt, trims built-in tool descriptions, and makes TodoListMiddleware opt-in. It also introduces new configurability for overriding built-in middleware and optimizes filesystem tools with paginated reads and streamed grep outputs.
Bears on: Agent capability
View sourceLangSmith introduced multiple platform updates, including control plane API support for assigning fixed resource tiers to deployments and IAM authentication support for Clustered Azure Redis. The release also updated LangSmith Model Context Protocol (MCP) tools to accept project UUIDs when querying run or thread histories.
Bears on: Deployment / data residency
View sourceLangChain and NVIDIA launched the NemoClaw for LangChain Deep Agents blueprint, a reference architecture for enterprise agent systems. The stack combines LangChain Deep Agents Code, NVIDIA's Nemotron 3 Ultra model, and the NVIDIA OpenShell runtime to optimize agent performance and inference efficiency.
Bears on: Integrations
View sourcePricing
From $39/seat/mo · free tier
hybrid
Included quota
OSS framework + LangGraph: free (MIT, self-host). LangSmith Developer (free): 1 seat, 5,000 base traces/mo, 14-day retention, basic evals, prompt playground, community support. Plus ($39/seat/mo): unlimited (purchasable) seats, 10,000 base traces/mo, 400-day retention available, full evals, custom dashboards, prompt hub, email support, and 1 free dev-sized managed deployment. Enterprise: SSO, HIPAA BAA, SOC 2 Type II, custom retention, self-hosting/VPC, higher volumes, dedicated support.
What is public
LangChain (langchain.com) ships the free open-source framework (MIT, 134,000+ GitHub stars; LangGraph OSS also free) plus the commercial LangSmith platform (observability, evals, deployment). LangSmith tiers: Developer (free, 1 seat, 5,000 traces/mo, 14-day retention), Plus ($39/seat/mo, unlimited seats, 10,000 base traces/mo, 400-day retention available, PAYG trace overage + 1 free dev deployment), and custom Enterprise (SSO, HIPAA BAA, SOC 2, self-hosting, dedicated support). A discounted Startup plan exists. LLM API usage is billed separately by your model provider.
Billing mechanics
Two cost components on LangSmith: per-SEAT licensing ($39/seat/mo on Plus; $0 for 1 Developer seat) + USAGE (traces and managed deployments). Each tier includes a base trace allowance (5,000 Developer / 10,000 Plus); beyond that you pay per 1,000 traces on PAYG (no hard cap on Plus) - ~$2.50/1,000 base (14-day) and ~$5.00/1,000 extended (400-day); some sources cite $0.50/1,000. LangSmith Deployment (managed LangGraph/'Fleet' hosting) bills by deployment runs + uptime - Plus includes 1 free dev-sized deployment; extra deployments are metered (~$0.005/run + uptime ~$0.0007/min dev, ~$0.0036/min prod). LLM provider usage is always separate.
Cost watchouts
Per-seat pricing scales linearly (a 10-person Plus team is $390/mo base before any trace overage); trace overage can balloon at production volume (millions of traces/mo -> four-figure overages, making Enterprise negotiation essential); the hosted LangGraph Platform/Deployment is NOT in the free tier (Plus-only, not clearly flagged pre-signup); LLM API costs (separate) typically dwarf the LangSmith fee (50-70% of TCO); Enterprise (SSO/SOC2/self-host) is custom - reported ~$2,000-5,000/mo+ and up
Variable cost rationale
Hybrid - a fixed per-seat base plus usage that scales with trace volume and managed-deployment runs/uptime; production trace volume (and any managed LangGraph deployments) drive the variable cost, and LLM provider usage (separate) scales on top
Additional watchouts
Premium-priced vs self-hostable alternatives (Langfuse is the frequently-cited cheaper, no-per-seat option); soft lock-in - LangSmith's deepest value is LangChain/LangGraph-native, and switching means re-instrumenting; hosted LangGraph Platform gated behind Plus (surprises some users); export trace/eval data before retention windows close; LangChain abstraction carries a documented cost premium for simple RAG/chatbot use cases
Overage / add-ons
Trace overage on Plus is pay-as-you-go per 1,000 traces with no hard cap: most detailed/recent sources cite ~$2.50/1,000 base (14-day) and ~$5.00/1,000 extended (400-day); a few cite $0.50/1,000 base. You can set workspace trace limits to cap spend. Managed deployments beyond the 1 free dev deployment bill by runs + uptime (~$0.005/run + ~$0.0007/min dev, ~$0.0036/min prod). Enterprise negotiates volume discounts.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free (Developer tier)
Lowest paid plan
Plus $39/seat/mo
Commercial notes
LangChain is the de facto OSS LLM framework (134,000+ stars); LangSmith is the 'flight data recorder' for LLM apps (tracing/eval/monitoring) with deepest LangChain/LangGraph integration (tracing via one env var); LangGraph adds agent orchestration; competes with Langfuse (MIT, self-hostable, no per-seat - the common cost-driven alternative), Arize Phoenix, Braintrust, W&B Weave, Datadog LLM Observability
Key ambiguities
Trace-overage rates and the exact trace DEFINITION vary across sources - the current official definition counts one end-to-end execution as a single trace (internal nodes/subgraphs not charged separately), but several 2026 write-ups describe per-step counting; base- vs extended-(400-day)-retention overage rates are quoted inconsistently ($0.50 vs $2.50 vs $5 per 1,000; the $2.50 base / $5 extended pair is the most-corroborated); Enterprise pricing is unpublished (~$2,000-5,000/mo+ reported)
Cancellation / refund
OSS is free (MIT, no contract); Developer is free; Plus is self-serve, seats billed monthly (pro-rated when added, no credit for removed seats), traces billed monthly in arrears; Enterprise is invoiced annually upfront; a discounted Startup plan offers generous free trace allotments
Support SLA / resale
Community support (Developer/OSS); email support on Plus; Enterprise adds SSO, HIPAA BAA, SOC 2 Type II, custom retention, self-hosted/hybrid (control-plane SaaS + data plane in your VPC) deployment, and dedicated support/SLAs; framework-agnostic tracing via OpenTelemetry SDKs (Python/TS/Go/Java)
Missing data
Enterprise pricing is unpublished (~$2,000-5,000/mo+ reported); trace-overage rates and the precise trace definition are quoted inconsistently across sources (base $2.50 / extended $5 per 1,000 is the most-corroborated). Seed 'Developer free; Plus $39/seat/mo + PAYG; OSS framework self-hostable' is accurate and current; LLM API usage is billed separately.
Related vendors
- Acrab — Singapore compute infrastructure company building a full stack…
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- AIsa — Unified resource and payment gateway for AI agents that lets them…
- AlphaBitCore — AI control plane that governs how models, agents, tools, and…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
Alternatives to LangChain
The closest documented capability profiles to LangChain among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Letta10.0 / 14A lighter documented profile than LangChain
- Agno12.5 / 14Adds documented Browser & Computer UseLangChain vs Agno →
- Haystack9.5 / 14A lighter documented profile than LangChainLangChain vs Haystack →
- Pydantic AI12.5 / 14Adds documented Browser & Computer UseLangChain vs Pydantic AI →
- Tray.ai11.5 / 14Fuller documented coverage on Security, Identity & Governance and Triggers & Channel Coverage
- Veecle9.5 / 14A lighter documented profile than LangChain
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded