Responses API Developer Client
Built for developers and prompt engineers who want direct access to `/v1/responses`, tool orchestration, and streaming internals.
OpenResponses sends requests directly from the client to the OpenAI Responses API. It exposes SSE event streams, raw request payloads, tool approval flows, and local conversation state so technical users can see what the model and tools are actually doing. This is the active product line that supersedes OpenAssistant.
What Makes OpenResponses Different
Direct Responses API path
Requests go straight to OpenAI's `/v1/responses` endpoint with user-owned API credentials instead of a proxy or older Assistants-style run orchestration layer.
SSE event visibility
Streaming text deltas, reasoning events, tool calls, and status updates are surfaced as typed client events rather than hidden behind a chat-style UI abstraction.
Tool execution with approvals
Computer use, web search, code interpreter, file search, and MCP flows are exposed with user approval points where the API or the app expects them.
Developer observability
Request inspection, token usage, event timelines, reasoning surfaces, and raw JSON make it useful for debugging integrations, not just chatting with a model.
Local credential handling
API keys and connection details stay in local storage such as the Keychain, with optional integrations added only when the user enables them.
Native file preparation
PDFKit and Vision are used on device to prepare local files and OCR content before it is packaged into Responses API payloads.
OpenAssistant successor
OpenResponses is where the active API-tooling work now lives. OpenAssistant remains as the older Assistants API line for reference, not as the current flagship client.
Technical Profile
Core APIs
- Primary: OpenAI Responses API over direct HTTPS
- Streaming: Server-Sent Events mapped into typed client events
- Tools: computer use, web search, code interpreter, file search, MCP
- Optional integrations: Notion, EventKit, Contacts, local network bridges
Data boundary
- Local: API keys, conversation state, request inspection, file preparation
- Remote: Requests and tool payloads sent directly to OpenAI when invoked
- Audience: developers, prompt engineers, technical users
- Status: active product line and successor to OpenAssistant
How OpenResponses Runs
Compose a direct request
The client builds a Responses API payload locally, including any user-selected tools, files, reasoning options, or prior response context.
Stream typed events
SSE events are parsed into client-side models so text deltas, reasoning phases, and tool transitions can be rendered in real time.
Approve or execute tools
Tool calls such as computer use, web search, code interpreter, and MCP can be approved, observed, and reconciled inside the native UI.
Inspect the outcome
The final response, raw payload details, and client-side logs are preserved locally so the user can audit what happened.