Preferences

The Model Preferences system allows per-chat model selection for agents with persistent storage in PostgreSQL. This enables dynamic model switching and preference management at runtime.

Model Types

export type ModelType = 'OPENAI' | 'GEMINI' | 'GROK' | 'DEEPSEEK';

Service Interface

class ModelPreferencesService {
  static async getModelPreference(
    agentId: string, 
    chatId: number
  ): Promise<ModelType | null>;

  static async setModelPreference(
    agentId: string, 
    chatId: number, 
    modelType: ModelType
  ): Promise<boolean>;

  static async deleteModelPreference(
    agentId: string, 
    chatId: number
  ): Promise<boolean>;
}

Database Schema

CREATE TABLE model_preferences (
  agent_id TEXT NOT NULL,
  chat_id INTEGER NOT NULL,
  model_type TEXT NOT NULL,
  created_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP,
  updated_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP,
  PRIMARY KEY (agent_id, chat_id)
);

CREATE INDEX idx_model_preferences_agent ON model_preferences(agent_id);
CREATE INDEX idx_model_preferences_chat ON model_preferences(chat_id);

Usage

Getting Model Preference

const modelType = await ModelPreferencesService.getModelPreference(
  'agent-123',
  456
);

if (modelType) {
  console.log(`Using preferred model: ${modelType}`);
} else {
  console.log('Using default model');
}

Setting Model Preference

const success = await ModelPreferencesService.setModelPreference(
  'agent-123',
  456,
  'GEMINI'
);

if (!success) {
  console.error('Failed to set model preference');
}

Removing Model Preference

await ModelPreferencesService.deleteModelPreference(
  'agent-123',
  456
);

Error Handling

The service implements graceful error handling:

• Returns null when no preference exists

• Returns false for failed operations

• Throws typed errors for database failures

• Validates input parameters

Best Practices

Data Access

• Use prepared statements

• Implement connection pooling

• Handle concurrent updates

• Validate input parameters

Performance

• Leverage database indexes

• Cache frequent lookups

• Batch operations when possible

• Monitor query performance

Related Documentation

• LLM Factory

• Database Schema

• Error Handling