Back to Gap Map Product / UX / Orchestration & agents

LangGraph

open core / Overall score: 5.0(leading)

Low-level orchestration framework from LangChain for stateful, multi-actor agent applications. It exposes an explicit graph and state-machine model - nodes, edges, persistence and human-in-the-loop checkpoints - rather than a higher-level abstraction, which is what makes fine-grained control over loops, branching and durable resumable execution possible.

The managed layer, LangSmith Deployment and LangSmith, is sold separately.

Openness

4 high confidence

4.0

license: MIT(core)
commercial: LangGraph-Platform/LangSmith
source: public
core features withheld: yes — the Agent Server runtime ships as the closed Elastic-2.0 langgraph-api package and self-hosting it needs LANGGRAPH_CLOUD_LICENSE_KEY

LangGraph's core is MIT-licensed and its source is public, but the Agent Server runtime that serves it ships as the separate langgraph-api package under the source-available Elastic License 2.0, and self-hosting it needs a license key verified over the network. That gated runtime keeps LangGraph short of fully open source.

https://docs.langchain.com/langsmith/deploy-standalone-server recorded 2026-09-29
The self-hosting docs list a license key for the standalone server: "LangSmith license key. This will be used to authenticate ONCE at server start up" and "required for license verification and usage reporting if not running in air-gapped mode".
https://github.com/langchain-ai/langgraph recorded 2026-06-17
MIT, ~35k stars, low-level agent orchestration
https://pypi.org/pypi/langgraph-api/json recorded 2026-09-29
The PyPI page of the langgraph-api Agent Server runtime package states "Elastic License 2.0 - see the" LICENSE file, a source-available non-OSI license.
https://raw.githubusercontent.com/langchain-ai/langgraph/main/LICENSE recorded 2026-09-29
The repository LICENSE file is the MIT License, granting "Permission is hereby granted, free of charge" for the LangChain, Inc. code, published openly on the main branch.

Adoption

5 high confidence

5.0

Adoption is measured by PyPI downloads of langgraph, though a large share is transitive, pulled in as a dependency of LangChain rather than installed directly, so the reading is directional rather than a precise count of direct users.

https://pypistats.org/api/packages/langgraph/recent recorded 2026-08-12
72,717,229 downloads in the trailing 30 days for langgraph

Capability

5 high confidence

5.0

Deepest durable-execution control among the orchestration peers.

https://docs.langchain.com/oss/python/langgraph/streaming recorded 2026-09-29
The LangGraph docs carry a dedicated page titled "Streaming - Docs by LangChain" under the Capabilities section.
https://docs.langchain.com/oss/python/langgraph/use-subgraphs recorded 2026-09-29
The LangGraph docs carry a dedicated subgraphs guide with the header link "/oss/python/langgraph/use-subgraphs" and the nav item titled "Subgraphs".
https://github.com/langchain-ai/langgraph recorded 2026-08-13
README - durable execution resuming from where a run left off, human-in-the-loop state inspection and modification, streaming, memory and persistence, branching and subgraphs; 39.6k stars, 6.7k forks
https://raw.githubusercontent.com/langchain-ai/langgraph/main/README.md recorded 2026-09-29
README describes durable execution and oversight: "automatically resuming from exactly where they left off" and "inspecting and modifying agent state at any point during execution", plus "Comprehensive memory".

Verified 2026-09-24