Configure Model Providers

Model provider configuration lives under Models in the admin panel. Administrators first create global models, then authorize them in team management. Built-in Clouisle provider identifiers include openai, openai_responses, anthropic, google, azure_openai, deepseek, moonshot, zhipu, qwen, baichuan, minimax, volcengine, siliconflow, xai, ollama, runway, pika, luma, kling, stability, midjourney, and custom.

Creating a Model

1. Select Add Model.
2. Select the provider and model type.
3. Fill in the display name (1-100 characters) and model_id (1-100 characters).
4. Fill in base_url and the API Key according to the provider; local providers can leave the API Key empty.
5. Set the context length, max output Tokens, default parameters, and capability flags.
6. Select Test Connection and save after it succeeds.

Model Types and Capabilities

Model types include chat, embedding, rerank, TTS, STT, audio generation, text-to-image, and text-to-video. Capability flags are stored in the capabilities JSON, for example vision, function calls, and streaming output. Before enabling tools for an Agent, confirm that the selected chat model supports function calls.

Endpoint Security

The Model Endpoint Whitelist in site settings matches the protocol, hostname, and port exactly by Origin and ignores the path. An empty list blocks all model endpoints. Do not write provider keys into documentation, frontend code, or Git.

Modifying a Model

Use the edit form and re-test when updating the API Key. The API only reports has_api_key and never echoes the key. Treat an updated model_id or type as a new model configuration, and re-check authorized teams and knowledge base index compatibility.