OpenAI Responses API Builder

The Responses API (launched March 2025) is OpenAI's unified interface for agentic AI tasks. It replaces the older Completions and Chat Completions for complex workflows, adding built-in tools (web search, file search, computer use), stateful runs, and streaming events. It's designed for multi-step agent tasks rather than single-turn Q&A.

Chat Completions is stateless (you send full history each time). Responses API maintains state with 'previous_response_id', has native built-in tools without extra code, supports background runs, and has a richer streaming event system. Use Chat Completions for simple Q&A; use Responses for agent workflows.

Built-in tools: web_search_preview (live web access), file_search (vector store search), computer_use_preview (screen/mouse/keyboard control), and code_interpreter. You can combine these with custom function tools. No extra setup needed — just include them in the tools array.

Pass the 'id' from a previous response as 'previous_response_id' in your next request. OpenAI maintains the conversation state server-side, so you don't need to resend the full history. This is more token-efficient for long agent sessions. Use 'store: false' if you don't want server-side storage.

GPT-4o, GPT-4o mini, o1, o3, o3-mini, and o4-mini support the Responses API. GPT-4.5 (preview) also works. Not all models support all built-in tools — computer_use requires specific models. Always check the model capabilities in the OpenAI documentation.