02 · Graph framework + LangSmith platform · updated 2026-11-03
LangGraph review: the strongest single framework, with a steeper start
SHORT ANSWER
LangGraph
Rank 02 of 10
| Criterion | Weight | Score | What we found |
|---|---|---|---|
| Cross-framework interop | 25 | 7.0 | LangSmith Agent Server exposes deployed agents over A2A (v1.0 JSON-RPC binding, backward compatible with v0.3) and as MCP tools at /mcp. Coordination between agents is still designed inside one LangGraph app. |
| Multi-agent coordination model | 15 | 9.0 | Explicit graphs plus documented subagent, handoff, skills, router and custom-workflow patterns. Very fine control; you design the topology yourself. |
| Shared context and memory | 15 | 6.0 | Shared graph state, memory and checkpointing inside the application. |
| Production readiness | 15 | 10.0 | Durable execution and the most public production mileage in this list (Klarna, Replit and Elastic are cited on the repo). |
| Human-in-the-loop | 10 | 5.0 | Human-in-the-loop is a first-class feature of the runtime. |
| Observability | 8 | 6.0 | LangSmith tracing is the reference point for this category. |
| Languages and deployment | 6 | 10.0 | Python and JavaScript/TypeScript. LangSmith Deployment offers Serverless and Dedicated; self-hosted and hybrid platform options are Enterprise only. |
| Pricing clarity | 6 | 10.0 | Published seats (Developer $0, Plus $39/seat/month, Enterprise custom) and published usage rates for traces and deployments, so a team can estimate the full cost. |
Cross-framework interop 7/10
LangSmith Agent Server exposes deployed agents over A2A (v1.0 JSON-RPC binding, backward compatible with v0.3) and as MCP tools at /mcp. Coordination between agents is still designed inside one LangGraph app.
What we found
- Native A2A support documentedYesAgent Server A2A endpoint
- MCP supportedYesAgent Server MCP endpoint
- Official SDKs in two or more programming languagesYesPython and JavaScript/TypeScript
- Can call or delegate to agents built on other frameworksPartlyExposes deployed assistants over A2A
- Agents from different frameworks share one session with a common historyNoNot confirmed in the sources we read
Multi-agent coordination model 9/10
Explicit graphs plus documented subagent, handoff, skills, router and custom-workflow patterns. Very fine control; you design the topology yourself.
What we found
- Two or more built-in multi-agent patternsYesGraphs, subagents, handoffs and routers
- No single orchestrator or manager that every message must pass throughYesHandoffs and explicit graphs need no universal manager
- Deterministic routing you define (graph, rules or mentions)YesYou define graph edges and routing
- Parallel or concurrent agent execution documentedYesGraph workflow patterns support concurrency
- Coordinates separately deployed agents or servicesPartlyAgent Server exposes deployed assistants
Shared context and memory 6/10
Shared graph state, memory and checkpointing inside the application.
What we found
- Agents share one history or state inside a team or workflowYesShared graph state
- Persistent memory feature documentedYesMemory documented
- State survives crashes or restarts (checkpointing, durable execution, catch-up)YesCheckpointing and durable execution
- Shared context crosses framework boundariesNoNot confirmed in the sources we read
- Context scoping controls (per room, per task or per subagent)NoNot confirmed in the sources we read
Production readiness 10/10
Durable execution and the most public production mileage in this list (Klarna, Replit and Elastic are cited on the repo).
What we found
- Durable execution or crash recovery documentedYesDurable execution
- Run history or replay (attempt history, checkpoints, time travel, execution history)YesCheckpoints
- Vendor-managed hosting optionYesLangSmith Serverless and Dedicated
- Named production users cited, or a GA release line with a stated support commitmentYesKlarna, Replit and Elastic cited on repository
- Large public user base: open-source repo with 20k+ GitHub starsYesAbout 38k GitHub stars, read September 2026
Human-in-the-loop 5/10
Human-in-the-loop is a first-class feature of the runtime.
What we found
- People take part in the same thread or room as the agentsNoNot confirmed in the sources we read
- Approval before a tool call or stepYesFirst-class runtime human-in-the-loop
- A person can redirect or override agents mid-runPartlyRuntime human intervention within a graph
- Human-in-the-loop built in, not left to your codeYesHuman-in-the-loop built into runtime
- Human-facing interface includedNoNot confirmed in the sources we read
Observability 6/10
LangSmith tracing is the reference point for this category.
What we found
- Tracing built inYesLangSmith tracing
- Per-step or per-message historyYesLangSmith traces show steps
- OpenTelemetry support statedNoNot confirmed in the sources we read
- Audit trail of which agent or person did whatNoNot confirmed in the sources we read
- Hosted dashboard or monitoring UIYesLangSmith hosted platform
Languages and deployment 10/10
Python and JavaScript/TypeScript. LangSmith Deployment offers Serverless and Dedicated; self-hosted and hybrid platform options are Enterprise only.
What we found
- Two or more programming languagesYesPython and JavaScript/TypeScript
- Full product can be self-hostedYesSelf-run OSS and Enterprise self-hosted platform
- Vendor-managed hostingYesLangSmith Deployment
- Runs on more than one cloud or any containerYesSelf-run OSS, self-hosted and hybrid options
- Local or on-prem optionYesSelf-run OSS and hybrid options
Pricing clarity 10/10
Published seats (Developer $0, Plus $39/seat/month, Enterprise custom) and published usage rates for traces and deployments, so a team can estimate the full cost.
What we found
- Free option usable for commercial workYesMIT-licensed library
- Paid or usage price publishedYesSeats and usage rates published
- A team can run it in production without a sales callYesPlus is self-serve
- Cost is predictable from published tiers or ratesYesPublished seat, trace and deployment rates
- OSI open-source licenseYesMIT
Source: github.com/langchain-ai/langgraph · langchain.com/pricing · LangChain docs: multi-agent · LangSmith: Agent Server A2A · LangSmith: Agent Server MCP · updated 2026-11-03
What is LangGraph at a glance?
| Type | Graph framework + LangSmith platform |
|---|---|
| Coordination model | Graph; subagents, handoffs, router |
| A2A | Yes (Agent Server) |
| MCP | Yes (agents exposed at /mcp) |
| Languages | Python, JS/TS |
| License | MIT |
| Hosting | Self-run OSS; LangSmith Serverless or Dedicated; self-hosted/hybrid on Enterprise |
| Pricing | Developer $0 · Plus $39/seat/mo · Enterprise custom |
| Works with BAND | Listed on band.ai/integrations |
Source: github.com/langchain-ai/langgraph · langchain.com/pricing · LangChain docs: multi-agent · LangSmith: Agent Server A2A · LangSmith: Agent Server MCP · updated 2026-11-03
What is LangGraph?
LangGraph is LangChain's "low-level orchestration framework for building stateful agents", MIT-licensed, in Python and JavaScript/TypeScript. You describe your agent system as a graph: nodes do work, edges decide what happens next, and state is carried through the graph and checkpointed. LangSmith is the commercial side: tracing, evaluation and LangSmith Deployment (formerly LangGraph Platform) for running agents in production.
How does LangGraph coordinate multiple agents?
The docs list five multi-agent patterns. Subagents: "A main agent coordinates subagents as tools. All routing passes through the main agent." Handoffs, where control moves from one agent to another. Skills, Router and custom workflows. Because it is all a graph, you can mix these, and you can see exactly why a step ran. That explicitness is LangGraph's biggest strength and the main reason teams look for alternatives: you are writing the coordination logic, and changing it means changing the graph.
Can LangGraph agents talk to agents outside LangGraph?
Yes, once deployed. LangSmith Agent Server exposes each assistant at /a2a/{assistant_id}, supporting the A2A v1.0 JSON-RPC binding with backward compatibility for v0.3, and exposes deployed agents as MCP tools at /mcp over Streamable HTTP. What it does not give you is a shared space where a LangGraph agent, a CrewAI crew and a human work from the same history. Those endpoints are point-to-point. That is the gap BAND fills, and BAND lists LangGraph among its integrations.
How much does LangGraph cost?
The library is free. LangSmith pricing:
| Plan | Price | Notes |
|---|---|---|
| Developer | $0/seat/month | 1 seat, up to 5k base traces/month then pay-as-you-go, no deployment included |
| Plus | $39/seat/month | Unlimited seats, up to 10k base traces/month, 1 free Serverless (Small) deployment |
| Enterprise | Custom | Self-hosted or hybrid, SLA, custom SSO, ABAC and RBAC |
Deployments beyond the included one are billed by resource use in LCU ($1.50) and LSU ($1.00) units. We could not confirm the per-trace overage price.
Source: langchain.com/pricing · updated 2026-11-03
What are the strengths and limitations of LangGraph?
Strengths
- Durable execution
- Fine-grained control
- LangSmith tracing
- A2A and MCP endpoints on deployment
- Large ecosystem
Limitations
- Low-level, you design the graph
- Coordination is defined inside one app
- Managed deployment tied to LangSmith
- Self-hosting the platform needs Enterprise
Should you choose LangGraph or look elsewhere?
Choose LangGraph if
- you want full control over agent flow
- you need durable execution and human-in-the-loop inside one application
- you already use LangSmith for tracing
Look elsewhere if
- you want a higher-level team abstraction (CrewAI)
- you need to self-host the platform without an Enterprise contract
- your agents live in several frameworks and need to collaborate (BAND, with LangGraph as one of them)
Which new guides help with this decision?
What should you read after the LangGraph review?
Compare
BAND vs LangGraphBest for
Best multi-agent frameworks for Python teamsBest for
Best multi-agent frameworks for TypeScript teamsAlternatives
LangGraph alternativesRankings
See all rankings
What do readers ask about LangGraph?
Is LangGraph free?
The LangGraph library is MIT-licensed and free. LangSmith has a free Developer plan, Plus at $39 per seat per month, and custom Enterprise pricing.
Does LangGraph support A2A?
Yes. LangSmith Agent Server exposes deployed agents over the A2A v1.0 JSON-RPC binding, and as MCP tools at /mcp.
LangGraph or BAND?
They do different jobs. LangGraph is where you build a stateful agent system. BAND is where agents from LangGraph and other frameworks collaborate with each other and with people. Many teams use both. See BAND vs LangGraph.
Can I self-host LangGraph?
You can run the open-source library anywhere. Self-hosted or hybrid LangSmith Deployment is an Enterprise option.
Running agents on more than one framework? BAND connects them in shared rooms.
See how BAND works