Integrate GigaChat with LangChain

LangChain GigaChat is a focused reference for using GigaChat inside LangChain, rather than calling the raw gigachat SDK directly: initializing and using the langchain-gigachat package, working with the chat model through LangChain message objects and runnables, generating embeddings via GigaChatEmbeddings, binding tools and getting structured outputs where the wrapper supports them, and wiring GigaChat into RAG pipelines using LangChain vector stores.

Use it specifically when the task is built with LangChain abstractions - chains, agents, or retrieval pipelines - on top of GigaChat. If the task doesn't actually need LangChain, the skill's own workflow says to prefer the gigachat-sdk-* skills instead of forcing a LangChain wrapper into the picture. It also stays out of two adjacent concerns by design: initial credential/TLS setup, which belongs to gigachat-setup and should be confirmed working first, and OpenAI-compatible base-URL routing, which belongs to gpt2giga.

GigaChat and GigaChatEmbeddings are initialized with the same authentication inputs as the raw SDK, but from there the recommended pattern is to work exclusively with LangChain message objects and runnables rather than raw GigaChat payloads, unless the task is specifically debugging the conversion layer between the two. For retrieval-augmented generation, the pattern is to pair GigaChatEmbeddings with a LangChain vector store and keep the retrieval logic itself inside LangChain rather than hand-rolling it. One practical gotcha called out explicitly: since both LangChain and the underlying SDK can implement retries, a naive combination can double-retry a failing request, so retry behavior needs to be checked rather than assumed. Two reference files carry further detail: references/chat-models.md for chat model and message/tool behavior, and references/embeddings-rag.md for embeddings and RAG patterns.

This skill sits alongside a small family of GigaChat-focused skills sharing the same source repository - gigachat-setup, the gigachat-sdk-* skills for direct SDK usage, and gpt2giga for OpenAI-compatible proxying - each scoped narrowly so the LangChain-specific guidance here doesn't have to re-explain setup or low-level SDK mechanics. Its default posture is to keep users inside LangChain abstractions unless there's an actual wrapper-boundary bug, calling out any place where langchain-gigachat's behavior doesn't have full parity with generic LangChain behavior rather than assuming it does.

Developers building LangChain-based chains, agents, or RAG pipelines who want GigaChat to be a drop-in chat/embeddings backend, and need to know where the LangChain wrapper's behavior around GigaChat specifically diverges from vanilla LangChain usage.