Generate Text with GigaChat SDK

GigaChat SDK Chat is a focused reference for direct Python integration with the official gigachat package's chat capabilities - client initialization via GigaChat(...), synchronous and asynchronous calls (chat, achat), and their streaming counterparts (stream, astream), plus message history and roles, model selection, token counting, and session caching through the X-Session-ID header.

Use it specifically when writing Python code that talks to the gigachat package directly, without LangChain or a proxy compatibility layer in between. Before using it, the skill's own workflow says to confirm setup is already solved - falling back to the gigachat-setup skill if not - and to prefer the official SDK over raw HTTP calls unless the user explicitly asks for REST. It deliberately stays out of two adjacent areas: custom tool/function calling, which belongs to gigachat-sdk-functions, and file or embedding handling, which belongs to gigachat-sdk-files-embeddings.

The recommended default is to start simple: for stateless chat, pass a plain string; only build an explicit Chat(messages=[...]) payload when conversation history or a richer message structure is actually needed. Streaming (stream/astream) is recommended for incremental UX, while async variants (achat/astream) should only be reached for when real concurrency is needed - not by default. Before large or cost-sensitive prompts, the skill calls for checking tokens_count first. Two reference files back these patterns with more depth: references/chat.md for request patterns and streaming, and references/models-and-tokens.md for model choice, token counting, context caching, and retries.

This skill is one piece of a small GigaChat skill family sharing the same source repository: gigachat-setup for initial configuration, gigachat-sdk-functions for tool/function calling, and gigachat-sdk-files-embeddings for files and embeddings - each intentionally out of scope here so the chat-focused guidance stays uncluttered. Session state is preserved across calls via the X-Session-ID header rather than resending full history every time.

Python developers building directly on top of the official gigachat SDK - rather than through LangChain or an OpenAI-compatible proxy - who want the simplest correct pattern for a given chat use case (sync vs. async, streaming vs. not, stateless vs. history-aware) instead of guessing at the SDK's API shape.