LangChainLangChain
LangChain 是 LLM 应用框架生态的主体（MIT，148k★）：模型/工具/检索的统一抽象层， 提供多类集成入口。同厂商三条线要分清：LangChain（组件抽象）、LangGraph（图编排）、Deep Agents（预制深 agent）。
LangChain is the main body of the LLM-app framework ecosystem (MIT, 148k stars): a unified abstraction layer for models/tools/retrieval with multiple integration entry points. Three lines from the same vendor to keep apart: LangChain (component abstraction), LangGraph (graph orchestration), Deep Agents (prebuilt deep agents).
适合与不适合When to use it, when not to
适合：要最大 provider/工具生态、要换模型不改代码、RAG 起步。 不适合：轻量直连单家 API；要开箱 agent 运行时。
Fits: the largest provider/tool ecosystem, swapping models without rewriting code, RAG starts. Not a fit: thin direct calls to one API; needing an out-of-the-box agent runtime.
中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-21
- 出品属地Origin
- 国外出品foreign-made LangChain, Inc.（美国）。LangChain, Inc. (US).
- 国内使用Direct access
- 大陆可直连direct OK langchain.com 与 python.langchain.com 直连实测 200（2026-10-21）。langchain.com and python.langchain.com measured 200 direct (2026-10-21).
- 备案状态CAC filing
- 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-21 检索）；开源项目未面向大陆公众提供服务，备案不适用。not in the CAC public registry (searched 2026-10-21); the open-source project does not serve the mainland public, so filing does not apply.
备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.
固定坐标系Fixed coordinate system
8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.
- 模型与开放条件Model & openness
模型接入层是它的看家能力：各家 LLM/嵌入/工具的统一接口族（provider 集成面巨大， 以官方 integrations 目录为准，未逐条清点）；它本身不调模型，是「换模型不改代码」的抽象层。
The model-access layer is its home-turf capability: a unified interface family across LLMs/embeddings/tools (the provider integration surface is huge, per the official integrations catalog, not itemized one by one); it does not call models itself — it is the abstraction layer where "switch providers without changing code".
- 运行位置Where it runs
Python/JS 库（pip/npm 装进你的进程）；GitHub 148,689★（API 核验 2026-10-21）—— LLM 应用框架生态的主体。配套 LangGraph（编排）/LangSmith（观测）是同一厂商的相邻产品线。
Python/JS libraries (pip/npm installed into your process); GitHub 148,689★ (API verified 2026-10-21) — the main body of the LLM application framework ecosystem. The companion LangGraph (orchestration) / LangSmith (observability) are adjacent product lines from the same vendor.
- 本地文件Local files
库形态跑在你自己的代码与进程里；数据落点由你的集成决定（向量库/文件都在你侧）。
Library form running inside your own code and process; where data lands is decided by your integrations (vector stores/files all on your side).
- 关机后的任务Tasks after shutdown
库本体无常驻语义；常驻与否取决于你用它写出的应用。官方托管线（LangSmith 平台）是另一产品面。
The library itself has no resident semantics; residency depends on the application you build with it. The official hosted line (the LangSmith platform) is a separate product surface.
- 工具与扩展Tools & extensions
工具定义与调用抽象是核心面；MCP 适配器线存在（langchain-mcp-adapters），支持细节未逐条核验。
Tool definition and invocation abstractions are a core surface; an MCP adapter line exists (langchain-mcp-adapters), support details not itemized.
- 上下文与记忆Context & memory
记忆/检索组件族（buffer/window/entity 等）是它的传统叙事；跨会话记忆要自己接存储——组件给模式不给托管。
The memory/retrieval component family (buffer/window/entity etc.) is its traditional narrative; cross-session memory requires wiring your own storage — the components give the pattern, not hosting.
- 权限与限制Permissions & limits
无账号体系（库形态）；权限面=你接的各家 API key 与工具能力。观测线（LangSmith）有数据上传面——默认行为未逐条核验。
No account system (library form); the permissions surface = the various vendors' API keys and tool capabilities you wire in. The observability line (LangSmith) has a data-upload surface — default behavior not itemized.
- 适合什么任务Best for
适合：要最大集成生态、要「换 provider 不改代码」的抽象层、RAG/工具流起步。 不适合：要轻量直连单家 API（抽象层是成本）；要开箱 agent 运行时（那是 LangGraph/Deep Agents 的线）。
Fits: wanting the largest integration ecosystem, wanting an abstraction layer where "switch providers without changing code", starting out with RAG/tool flows. Does not fit: lightweight direct calls to a single vendor's API (the abstraction layer is a cost); wanting a batteries-included agent runtime (that is the LangGraph/Deep Agents line).
头号误解Top misconception
- 把 LangChain/LangGraph/Deep Agents 混成一团——同厂商三条线：组件抽象层、图编排、预制深 agentBlurring LangChain/LangGraph/Deep Agents into one blob — three lines from one vendor: component abstraction, graph orchestration, prebuilt deep agents
- 把 148k★ 的生态当「每个集成都一等公民」——第三方集成质量参差，看官方维护标记Reading 148k stars as 'every integration is first-class' — third-party integration quality varies; check official-maintenance markers
- 用观测线不查数据上传面——LangSmith 上传 trace 是产品语义，接前先核默认行为Plugging in the observability line without checking its upload surface — trace uploads are product semantics for LangSmith; verify defaults first
未知项清单Known unknowns
- integrations 全目录与官方维护标记（未逐条清点）
- MCP 适配器当前能力面（未逐条核验）
- LangSmith 数据上传默认行为细节（未逐条核验）
- The full integrations catalog and official-maintenance markers (not itemized one by one)
- The current capability surface of the MCP adapter (not itemized)
- Details of LangSmith's default data-upload behavior (not itemized)
证据来源Evidence sources
判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30，内容更新日content updated on 2026-10-21。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| repo | langchain-ai/langchain · 仓库（MIT；148,689★，最近推送 2026-10-29，GitHub API 核验）langchain-ai/langchain · repository (MIT; 148,689★, last push 2026-10-29, GitHub API verified) | https://github.com/langchain-ai/langchain |
| docs | LangChain · 官方文档LangChain · official docs | https://python.langchain.com |
| changelog | LangChain · 官方 changelogLangChain · official changelog | https://changelog.langchain.com |
| changelog | langchain-ai/langchain · Releases（changelog）langchain-ai/langchain · Releases (changelog) | https://github.com/langchain-ai/langchain/releases |
实测记录Measured runs
本档未跑集成与性能实测；结论只覆盖 API 与官方文档披露的能力面。
This site has not completed measured runs. The test protocol is at protocols/framework/_protocol.md.
相关条目Related entries
- LangGraphLangGraph同厂商编排线：LangChain 给组件，LangGraph 给图与运行时。Same-vendor orchestration line: LangChain gives components, LangGraph gives the graph and runtime.
- Deep AgentsDeep Agents同厂商预制线：不想手搭全套循环时用它的深 agent 骨架。Same-vendor batteries-included line: when you don't want to hand-build the whole loop, use its deep-agent skeleton.
- CursorCursor另一层 · AI 编程Other layer · AI coding
- Claude Code IDE 扩展Claude Code (IDE extension)另一层 · AI 编程Other layer · AI coding
本页由This page is generated from the ai-handbook 数据层生成（CC BY 4.0）。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.