Can I use the OpenAI Responses API with Claude, Gemini and other models?
Cognocient AI supports POST /v1/responses, including streaming and function tools, on every provider. Budgets, guardrails, routing, caching and cost tracking apply exactly as they do to chat completions.
Cognocient AI exposes POST /v1/responses alongside /v1/chat/completions, so code written against OpenAI's newer Responses API works with a one-line base-URL change.
Streaming works the same way (stream=True), emitting the standard response.output_text.delta, response.function_call_arguments.delta and response.completed events.
How it works
Cognocient AI translates the Responses request into a chat request, runs it through the same pipeline as /v1/chat/completions, and translates the result back. That means everything you already configured applies with no extra setup:
- Attribution headers (
X-Cost-Feature,X-Cost-User, …) and cost tracking - Budgets, the velocity circuit breaker and the Emergency Freeze
- PII / secret / prompt-injection guardrails
- The Auto Router, routing rules and load-balanced groups
- Semantic caching, provider failover and rate limits
- Every provider, not just OpenAI
Calls appear in the dashboard like any other call.
Supported
| Feature | Notes |
|---|---|
input as text or an item list | Messages, function_call and function_call_output items |
instructions | Sent as the system message |
| Streaming | Full Responses event stream |
Function tools and tool_choice | Including parallel tool calls |
| Structured output | text.format with json_schema or json_object |
| Image input | input_image with an image_url |
max_output_tokens, temperature, top_p, reasoning.effort | Mapped to the provider's equivalent |
Not supported
Cognocient AI's Responses endpoint is stateless: it keeps no response history. These return a 400 with code: "unsupported_parameter":
previous_response_id,conversation: send the full conversation ininputinsteadbackgroundmode and storedprompttemplates- Hosted tools that run on OpenAI's servers:
web_search,file_search,code_interpreter,computer_use, remotemcp - File-ID inputs (
input_imageneeds a URL,input_fileis not supported)
store and metadata are accepted and ignored. Responses are never stored.
Calls made through /v1/responses are recorded with the same endpoint label as chat completions, so they are not separated in endpoint breakdowns.
Related articles
Audio & Rerank Endpoints
Transcription, translation, speech and rerank through Cognocient AI, with the same budgets, guardrails and cost tracking as chat.
OpenAI Batch API
Create batches through Cognocient AI and have each batch's discounted cost attributed to your tags when it finishes. Growth plan.
MCP Gateway
Govern MCP servers: tool allow/deny policy, argument scanning, per-call pricing and full attribution. Business plan.