Bestony AI Provider for Xiaomi MiMo

No. It requires the PHP AI Client SDK, which is provided by WordPress 7.0 or by the AI plugin. The provider stays silent when the SDK is missing.

The API key (on Settings → Connectors) and, if you change them from the defaults, the settings on Settings → Bestony AI Provider for Xiaomi MiMo. Both can instead be set with an environment variable or constant.

Because the Connectors screen and the connectors_ai_xiaomimimo_api_key option are WordPress 7.0 features. The SDK's own credential lookup reads only environment variables and constants, so on 6.9 you could paste a key into a UI that does not exist yet. WordPress 7.0 wires the stored key to the provider for you.

The standard API (https://api.xiaomimimo.com/v1) for a pay-as-you-go key, which is billed against your account balance. If you bought a Token Plan subscription, use the cluster that matches where the plan was issued — China, Singapore or Europe — because Token Plan keys and their quota are separate from pay-as-you-go keys.

mimo-v2.6-pro, mimo-v2.6-flash, mimo-v2.6-pro-ultraspeed and mimo-v2.5. The model list endpoint reports no capabilities, so this list lives in the plugin source. If your model supports images but is not on the list, set XIAOMIMIMO_MODEL_INPUT_MODALITIES=text,image.

MiMo implements OpenAI JSON mode only (response_format: {"type":"json_object"}) and rejects a JSON schema, so this plugin sends json_object and drops the schema. MiMo guarantees valid JSON but not the exact fields, so the model relies on the prompt for the shape. Try setting Structured output to Send no response_format if a feature is unhappy, or point that feature at another provider.

MiMo forces its own temperature/top_p while thinking is enabled, which is the default. Set Thinking mode to Disable thinking (or XIAOMIMIMO_THINKING=disabled) to make those parameters apply.

MiMo's open platform serves text/multimodal-understanding and speech models, not text-to-image, so the plugin declares no image-generation capability.

The PHP AI Client exposes text, image, speech, video and embedding entry points, and the MiMo speech endpoints do not fit the shape the SDK sends. The ASR/TTS models still appear in the model list, but with no capability they cannot be selected.

In the WordPress options table on your own site, under the AI Client's connector option (connectors_ai_xiaomimimo_api_key), or in an environment variable or PHP constant if you set one. It is never transmitted anywhere except to the Xiaomi MiMo host configured above.

Only what you send to the AI: your prompts, and whatever the calling plugin or theme adds to them (system instructions, conversation history, tool definitions, attached files). See External services below for the exact endpoints.

Yes: Settings → Bestony AI Provider for Xiaomi MiMo, which chooses the API host, the thinking mode and the structured-output mode. The API key still lives on Settings → Connectors.