> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.asi1.ai/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.asi1.ai/_mcp/server. # OpenAI Compatibility > Use ASI:One with the OpenAI client libraries - supported parameters, the two endpoints, and ASI:One's own parameters ## Overview ASI:One implements both of OpenAI's request formats, so you can use the existing OpenAI client libraries and change only the base URL: * **[`/v1/chat/completions`](/documentation/build-with-asi-one/chat-completions)** - the Chat Completions format. The default, and what every example in these docs uses unless it says otherwise. * **[`/v1/responses`](/documentation/build-with-asi-one/responses)** - the Responses format, including retrieving, chaining, cancelling and deleting a response by id. Each endpoint supports its own OpenAI parameter set. Where they differ, the sections below say which endpoint to use. ## API compatibility ### Standard OpenAI parameters These take the same values on `/v1/chat/completions` as they do on OpenAI's API: | Parameter | What it does | | ---------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------- | | `model` | The ASI:One model to use. See [Models](/documentation/models). | | `messages` | The conversation so far. | | `temperature` | Sampling temperature, `0` to `2`. | | `top_p` | Nucleus sampling probability mass. | | `max_tokens` | Upper bound on the tokens generated. | | `stop` | Up to four strings that end generation when produced. | | `stream` | Stream the reply as server-sent events. | | `tools`, `tool_choice` | See [Tool Calling](/documentation/build-with-asi-one/tool-calling). | | `parallel_tool_calls` | Whether the model may call several tools in one turn. | | `response_format` | See [Structured Data](/documentation/build-with-asi-one/structured-data). | | `reasoning_effort` | How deeply the model reasons, on `/v1/chat/completions`. See [Reasoning](/documentation/build-with-asi-one/reasoning#choosing-a-reasoning-depth). | | `web_search_options` | Gives the model live web search, on `/v1/chat/completions`. See [Web Search](/documentation/build-with-asi-one/web-search). | On `/v1/responses` the same set applies, with `input` in place of `messages` and `max_output_tokens` in place of `max_tokens`, plus `store`, `metadata`, `background` and [`reasoning`](/documentation/build-with-asi-one/reasoning). `stop`, `reasoning_effort` and `web_search_options` belong to `/v1/chat/completions` alone. On `/v1/responses` the depth goes in the nested `reasoning.effort`, and a top-level `reasoning_effort` there is rejected with a `400`; search is the hosted `web_search` tool instead. See [Responses API](/documentation/build-with-asi-one/responses). Structured output is the one parameter that changes shape between the two: `response_format` on `/v1/chat/completions`, `text.format` on `/v1/responses`. [Structured Data](/documentation/build-with-asi-one/structured-data#on-the-responses-api) shows both. > **Tip** > > Each model is tuned for the sampling values it was evaluated at, so send > `temperature` and `top_p` only when you have a reason to move them. See > [Sampling defaults](/documentation/models#sampling-defaults) for the values > and for when deviating helps. ### Unrecognized parameters A request carrying a top-level field the endpoint does not recognize is rejected with a `400`, and the error names the field: ```json { "error": { "message": "unknown parameter: max_token", "type": "invalid_request_error", "code": "unknown_parameter", "param": "max_token", "status": 400 } } ``` This is deliberate. A misspelled `max_token` fails immediately and tells you what to fix, rather than being dropped and leaving you to work out why the reply ignored your limit. In [planner mode](/documentation/build-with-asi-one/planner) the planner manages its own model calls and takes `temperature` alone. ## Custom parameters These are specific to ASI:One and have no OpenAI equivalent. | Parameter | Endpoint | What it does | | ----------------- | ---------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- | | `planner_mode` | `/v1/chat/completions` | Routes the request to the planner. See [Planner Mode](/documentation/build-with-asi-one/planner). | | `agents` | `/v1/chat/completions` | Agentverse addresses the planner may use. See [Planner Mode](/documentation/build-with-asi-one/planner). | | `enable_thinking` | Both | Whether the model reasons first. See [Reasoning](/documentation/build-with-asi-one/reasoning). | | `thinking_budget` | Both | The older way to size reasoning, superseded by `reasoning_effort`. See [Reasoning](/documentation/build-with-asi-one/reasoning#legacy-depth-control). | All four are top-level fields. With the OpenAI SDK, pass them through `extra_body` so they land at the top level of the request body. `reasoning_effort` is not one of them: it is a standard OpenAI field, so the SDK sends it directly. ## Next steps 1. **[Chat Completions API](/documentation/build-with-asi-one/chat-completions)** - The default endpoint in full, including streaming 2. **[Responses API](/documentation/build-with-asi-one/responses)** - The stateful endpoint, with retrieve and chaining 3. **[Reasoning](/documentation/build-with-asi-one/reasoning)** - The ASI:One parameters alongside OpenAI's `reasoning` object 4. **[Planner Mode](/documentation/build-with-asi-one/planner)** - Multi-step work against Agentverse agents and tools