openai-assistants

Build stateful chatbots by creating an assistant, starting a thread, adding user messages, executing runs, and polling for completion. Assistants API v2 persists conversation state across multiple turns, supporting file attachments, code execution, semantic search, and function calling without managing message history manually.

Threads are persistent conversation containers holding up to 100k messages. Runs execute assistant logic against a thread asynchronously, orchestrating tools like code_interpreter and file_search. Together they enable multi-turn stateful interactions without rebuilding context between requests.

Yes. The Assistants API supports file_search with vector stores for RAG-enabled document retrieval across thousands of documents. Upload files via batch indexing, attach vector stores to assistants, and the API performs semantic search without manual embedding or retrieval logic.

Runs execute asynchronously. Use polling to check run status until completion, or consume server-sent events (SSE) for streaming results. The API handles background processing, eliminating timeout concerns for code execution and file operations.

Yes. The code_interpreter tool executes Python code with file I/O. File attachments and vector stores enable document uploads, batch indexing, and retrieval. Both features work natively within the run execution model without external infrastructure.

The Assistants API is sunsetting in favor of the Responses API. This Skill includes migration guidance for transitioning stateful chatbot logic, threads, runs, and tools to the newer API without rebuilding conversation persistence.