OpenAI Assistants API provides stateful, tool-enabled AI with persistent threads. Key concepts: Assistants define behavior (instructions, model, tools), Threads maintain conversation history and context, Messages belong to threads, Runs execute assistant logic. File search enables RAG over uploaded documents without building your own vector store - upload PDFs, code files, or any text-based documents and the assistant searches them automatically. Code interpreter runs Python in sandboxed environment for data analysis, chart generation, and file processing. Function calling: define tools as JSON schemas, assistant calls them when needed, your app executes and returns results. Production considerations: implement webhook or polling for run status, handle requires_action state for function calls, implement thread pruning to manage costs, use streaming for real-time responses. Store thread IDs per user for conversation continuity.