LLM Integration in XSky

This skill provides knowledge for LLM provider integration in XSky.

Supported Providers

Configuration

const xsky = new XSky({
  llms: {
    default: {
      provider: "openai",
      model: "gpt-4o",
      apiKey: process.env.OPENAI_API_KEY,
      config: {
        temperature: 0.7,
        maxTokens: 4000
      }
    },
    fast: {
      provider: "anthropic",
      model: "claude-3-haiku-20240307",
      apiKey: process.env.ANTHROPIC_API_KEY
    }
  },
  planLlms: ["default"],     // For workflow planning
  compressLlms: ["fast"]     // For memory compression
});

Streaming Callbacks

callback: {
  async onMessage(message) {
    switch (message.type) {
      case "text": // Streaming text
      case "tool_use": // Tool being called
      case "tool_result": // Tool result
      case "agent_start": // Agent started
      case "agent_result": // Agent completed
      case "finish": // Task done
    }
  }
}

Human-in-the-Loop

callback: {
  async onHumanConfirm(context, prompt) {
    return await showConfirmDialog(prompt);
  },
  async onHumanInput(context, prompt) {
    return await showInputDialog(prompt);
  },
  async onHumanSelect(context, prompt, options, multiple) {
    return await showSelectDialog(prompt, options);
  }
}

Token Optimization

1. Use compressThreshold to auto-compress long conversations
2. Use efficient models for simple tasks
3. Truncate page content before sending to LLM
4. Use progressive disclosure in prompts

Adding New Provider

1. Install AI SDK adapter: pnpm add @ai-sdk/<provider>
2. Create provider file in src/llm/providers/
3. Register in provider factory
4. Update LLMProvider type

Key Source Files