@cloudbase/agent-adapter-llm

Direct LLM adapter for AG-Kit agents with support for OpenAI and Anthropic models.

Features

• ✅ Direct OpenAI and Anthropic model integration
• ✅ Automatic streaming message conversion to AGUI events
• ✅ Support for tool calls (function calling)
• ✅ Type-safe with TypeScript
• ✅ Observable-based event streaming
• ✅ No additional abstraction layers

Installation

npm install @cloudbase/agent-adapter-llm openai @anthropic-ai/sdk

Quick Start

With OpenAI

import OpenAI from 'openai';
import { LLMAgent } from '@cloudbase/agent-adapter-llm';

const openai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY
});

const agent = new LLMAgent({
  agentId: 'my-agent',
  model: openai,
  modelName: 'gpt-4o-mini',
  systemPrompt: 'You are a helpful assistant.',
  temperature: 0.7,
  maxTokens: 2000
});

const observable = agent.run({
  messages: [
    { id: 'msg_1', role: 'user', content: 'Hello!' }
  ],
  runId: 'run_123',
  threadId: 'thread_456',
  tools: []
});

observable.subscribe({
  next: (event) => {
    console.log('Event:', event);
  }
});

With Anthropic

import Anthropic from '@anthropic-ai/sdk';
import { LLMAgent } from '@cloudbase/agent-adapter-llm';

const anthropic = new Anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY
});

const agent = new LLMAgent({
  agentId: 'my-agent',
  model: anthropic,
  modelName: 'claude-3-5-sonnet-20241022',
  systemPrompt: 'You are a helpful assistant.',
  temperature: 0.7,
  maxTokens: 2000
});

AGUI Events

The agent emits the following AGUI protocol events:

• RUN_STARTED - Run has started
• TEXT_MESSAGE_START - Text message streaming started
• TEXT_MESSAGE_CONTENT - Text content delta
• TEXT_MESSAGE_END - Text message streaming ended
• TOOL_CALL_START - Tool call started
• TOOL_CALL_ARGS - Tool call arguments delta
• TOOL_CALL_END - Tool call ended
• RUN_FINISHED - Run completed successfully
• RUN_ERROR - Run encountered an error

Configuration

LLMAgentConfig

interface LLMAgentConfig {
  agentId: string;           // Unique agent identifier
  name?: string;             // Agent name
  description?: string;      // Agent description
  model: OpenAI | Anthropic; // Model provider instance
  modelName: string;         // Model name (e.g., 'gpt-4o-mini', 'claude-3-5-sonnet-20241022')
  systemPrompt?: string;     // System prompt
  temperature?: number;      // Temperature (0-2)
  maxTokens?: number;        // Max tokens to generate
  threadId?: string;         // Thread ID
}

Streaming Example

import { EventType } from '@ag-ui/client';

observable.subscribe({
  next: (event) => {
    switch (event.type) {
      case EventType.TEXT_MESSAGE_CONTENT:
        process.stdout.write(event.delta);
        break;
      case EventType.TOOL_CALL_START:
        console.log(`\nTool: ${event.toolCallName}`);
        break;
      case EventType.RUN_FINISHED:
        console.log('\nDone!');
        break;
    }
  }
});

Tool Calling

const tools = [
  {
    name: 'get_weather',
    description: 'Get the current weather',
    parameters: JSON.stringify({
      type: 'object',
      properties: {
        location: { type: 'string' }
      },
      required: ['location']
    })
  }
];

agent.run({ messages, runId, threadId, tools });

License

MIT