AI Models and Presets

The AI models module centrally manages the large-model configurations the backend calls, and provides a set of presets to choose from.

Configuring a Model

Field	Description
`alias`	Unique alias; only letters, digits, underscores, and hyphens are allowed (`^[a-zA-Z0-9_-]+$`)
`modelName`	Model name, e.g. `deepseek-v4-flash`
`apiUrl`	Endpoint URL
`apiKey`	Access key
`description`	Description
`isDefault`	Whether this is the default model

Marking a model as default automatically clears the default flag on other models, so there is always at most one default model.

API Key Masking

When model information is returned, the API key is masked:

Length under 12: returns ****.
Otherwise returns a value shaped like ab****yz.

On update, the masked value is not written back to the database, so the real key is not overwritten.

Query Endpoints

Method	Path	Description	Auth
GET	`/api/ai-models/default`	Default model (no key)	Login
GET	`/api/ai-models/by-alias/:alias`	Query by alias (no key)	Login
GET	`/api/ai-models`	List (`search`, pagination)	`ai_models`
POST	`/api/ai-models`	Create	`ai_models.create`
GET	`/api/ai-models/:id`	Details	`ai_models`
PUT	`/api/ai-models/:id`	Update	`ai_models.edit`
DELETE	`/api/ai-models/:id`	Delete	`ai_models.delete`
POST	`/api/ai-models/test`	Connection test	`ai_models`

The public information returned by default and by-alias does not include the API key, so it is safe to use at runtime to fetch a model by alias.

Presets

Presets are an optional list of models, divided into groups such as the built-in DeepSeek and Xiaomi MiMo (小米 MiMo):

Group	Preset aliases	Default URL
DeepSeek	`deepseek-v4-flash`, `deepseek-v4-pro`	`https://api.deepseek.com/v1/chat/completions`
Xiaomi MiMo (小米 MiMo)	`mimo-v2.5`	`https://api.xiaomi.com/v1/chat/completions`

Preset fields include group, alias, model name, URL, description, whether active (isActive), and sort order. Once enabled, they can be displayed ordered by sortOrder for the frontend’s model selector.

Preset create/update/delete accepts either family of permissions, ai_models.* or ai_models.presets.* (meeting either is enough).

GET /api/ai-models/presets list (supports search, group)
GET /api/ai-models/presets/groups groups
GET /api/ai-models/presets/active active presets (login required)
POST / PUT / DELETE create/update/delete

Connection Test

POST /api/ai-models/test sends one minimal chat request to the target URL (max_tokens: 10, temperature: 0), with a 30-second timeout, and returns:

{ "success": true, "message": "...", "responseTime": 820, "model": "deepseek-v4-flash" }

The request body can pass modelId to reuse an already-saved URL and key, or pass the three fields apiUrl, apiKey, and modelName directly.