Build resilient agents.
37k
Stars
6.2k
Forks
613
Open issues
100+
Contributors
AI Analysis
LangGraph is a low-level orchestration framework for building stateful, long-running agents with durable execution, human-in-the-loop capabilities, and comprehensive memory management. It serves developers and enterprises building production AI agents that require resilience, observability, and complex state management—not a general-purpose tool for casual LLM experimentation, but specialized infrastructure for agent systems at scale.
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
LangGraph brings stateful, durable graph-based orchestration to production AI agents
LangGraph is a low-level Python (and JS) framework for building stateful, long-running agents using directed graph structures. It targets developers who need durability, human-in-the-loop workflows, and persistent memory beyond what simple LLM chains provide. Named users include Klarna, Replit, and Elastic. It occupies the orchestration layer between raw LLM calls and full application frameworks, integrating tightly with LangChain and LangSmith for observability and deployment. Its graph model gives explicit control over execution flow, state transitions, and error recovery.
Created in August 2023 as LangChain's answer to complex, multi-step agent orchestration. Emerged during a period when ReAct loops and simple chains proved insufficient for production agents needing persistence and branching.
Growth was propelled by the broader LLM agent wave of 2023-2025, LangChain's large existing user base funneling developers toward the framework, and genuine pain around agent reliability in production. 717 new stars in the last 7 days as of June 2026 suggests sustained organic interest rather than a fading trend. The companion LangGraph.js repo and sibling projects like langgraph-swarm-py indicate ecosystem expansion.
README names Klarna, Replit, and Elastic as production users. A 'built-with-langgraph' case studies page is linked, suggesting documented external deployments. PyPI download badge is prominently displayed, implying substantial download volume. LangSmith Deployment integration targets production-scale, long-running workflows. These signals point to meaningful production adoption, though independent third-party verification is not available from the README alone.
Appears to use a directed graph abstraction where nodes represent computation steps and edges represent conditional or unconditional transitions. State is likely managed as a typed schema passed through nodes, enabling checkpointing and resumption. Durable execution likely relies on a persistence layer (checkpointer) that snapshots state after each node. The framework appears to be modular, allowing standalone use or integration with LangChain components.
not documented in README
Last push was June 23, 2026 — one day before the evaluation date — indicating extremely active, ongoing development. The 5,956 forks and 35,553 stars, combined with daily commits, suggest a healthy, well-resourced maintenance cadence consistent with commercial backing from LangChain, Inc.
ADOPT IF: you are building production agents that require durable execution, human-in-the-loop checkpointing, stateful memory across sessions, or complex branching workflows — especially if already in the LangChain ecosystem. AVOID IF: your use case is simple single-turn LLM calls, you have a strong preference for LLM-agnostic tooling with no vendor ecosystem ties, or you need proven general-purpose workflow durability that a tool like Temporal already provides. MONITOR IF: you are evaluating agent frameworks but uncertain whether the graph abstraction fits your team's mental model, or if you are watching how LangSmith Deployment's commercial layer evolves relative to open-source capabilities.
Independent dimensions
Mainstream potential
8/10
Technical importance
8/10
Adoption evidence
7/10
- Tight coupling to the LangChain/LangSmith commercial ecosystem may create friction for teams wanting full vendor independence, particularly around production deployment features.
- The graph programming model has a non-trivial learning curve; teams unfamiliar with state machine or DAG concepts may struggle to adopt it efficiently.
- Fast-moving API surface in a rapidly evolving space means migration burden between versions may be significant for production users.
- The framework's value proposition in durable execution overlaps with mature general-purpose workflow engines (Temporal, Prefect), which have broader operational tooling and longer production track records.
- Commercial pressure from LangChain, Inc. could push important features toward paid LangSmith tiers over time, narrowing the practical open-source experience.
LangGraph is likely to consolidate as the default orchestration substrate for LangChain-ecosystem agent builders, while continuing to face pressure from both higher-level frameworks (Deep Agents, CrewAI) and general-purpose durable workflow engines as the agent infrastructure space matures.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
Information
- Language
- Python
- License
- MIT
- Last updated
- 12h ago
- Created
- 36mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
StateGraph silently drops node output keys not declared in TypedDict
State channels are reference-transparent end-to-end (read AND write) — local_read()'s missing copy() is a symptom, not the root cause (LastValue et al. alias caller/node objects from the first .invoke())
feat: CCS Runtime Governance Integration — Formal Behavioral Conformance for Checkpointers
Similar repos
langchain-ai/deepagents
Deep Agents is an open-source agent framework that provides a production-ready...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
langchain-ai/langgraph
this
|
37k | +515 | Python | 8/10 | 12h ago |
|
langchain-ai/langgraphjs
|
3.1k | — | TypeScript | 8/10 | 1d ago |
|
langchain-ai/langchain
|
141.5k | — | Python | 8/10 | 4h ago |
 langgraph4j/langgraph4j
|
1.8k | — | Java | 8/10 | 5d ago |
|
langchain-ai/deepagents
|
26k | — | Python | 8/10 | 8h ago |
|
langchain-ai/chat-langchain
|
6.4k | — | TypeScript | 8/10 | 13h ago |
AutoGen focuses on multi-agent conversation patterns with strong Microsoft ecosystem ties. LangGraph offers more explicit graph control and durable execution primitives, making it preferable for workflows needing fine-grained state management and resumption after failure.
CrewAI provides a higher-level, role-based multi-agent abstraction with simpler onboarding. LangGraph is lower-level and more flexible but requires more boilerplate. CrewAI may suit rapid prototyping; LangGraph suits teams needing precise control over execution logic.
Temporal and Prefect are general-purpose durable workflow engines with broader ecosystem support. LangGraph is LLM-native, offering graph primitives, memory, and model-aware tooling that general workflow engines lack, but trading off some operational maturity.
LangChain handles LLM integrations and composable chains. LangGraph is the orchestration and state management layer on top. The two are complementary, not competing, though some developers may find LangChain chains sufficient for simpler workflows.
Deep Agents (langchain-ai/deepagents, 25k stars) is a higher-level framework built on LangGraph. For teams wanting opinionated structure and faster setup, Deep Agents may be preferable; LangGraph is the right choice when custom graph topology and fine-grained control are required.
