Skip to main content

AI Agent v3 Launch

· 3 min read
Eduardez
Eduardez
MoLOS Lead Developer

We're excited to announce the launch of AI Agent v3, a complete redesign of our AI architecture built on the Vercel AI SDK. This release brings modular architecture, improved streaming, and real-time progress tracking.

What's New in v3?​

AI SDK Integration​

The v3 agent is built on top of the Vercel AI SDK, providing:

  • Unified API for multiple LLM providers
  • Built-in streaming support
  • Type-safe interfaces
  • First-class tool support

Modular Architecture​

The agent is now split into focused modules:

agent/v3/
├── core/
│   ├── molos-agent.ts       # Main agent orchestration
│   ├── config.ts            # Agent configuration
│   └── state.ts             # State management
├── multi-agent/
│   ├── module-registry.ts   # Module registration
│   ├── delegation-tools.ts  # Module delegation
│   └── module-spec.ts       # Module specification
├── providers/
│   ├── factory.ts           # Provider factory
│   └── index.ts             # Provider implementations
├── tools/
│   ├── schema-converter.ts  # Schema conversion
│   ├── tool-wrapper.ts       # Tool execution
│   └── index.ts             # Tool registry
└── types/
    └── index.ts             # TypeScript definitions

Key Features​

Real-Time Progress Streaming​

Watch your agent execute tasks in real-time with detailed progress updates:

interface ProgressEvent {
  type: 'step_start' | 'step_complete' | 'step_error' | 'planning' | 'complete';
  stepNumber: number;
  message: string;
  details?: Record<string, unknown>;
  timestamp: Date;
}

Modular Tool System​

The agent discovers and uses tools from registered modules automatically:

interface ModuleTool {
  name: string;
  description: string;
  parameters: Record<string, ToolParameter>;
  handler: (params: unknown) => Promise<ToolResult>;
}

interface ToolParameter {
  type: 'string' | 'number' | 'boolean' | 'array' | 'object';
  description: string;
  required: boolean;
  default?: unknown;
}

Multi-Provider Support​

Support for multiple LLM providers with automatic fallback:

const supportedProviders = [
  'openai',
  'anthropic',
  'google',
  'azure',
  'openai-compatible', // For Z.AI and others
  'custom'
];

Enhanced Error Handling​

Intelligent error recovery and user-friendly messages:

try {
  const result = await agent.execute(prompt);
} catch (error) {
  if (error instanceof ModuleExecutionError) {
    // Provide guidance for module errors
    showError(`Module ${error.moduleName} failed: ${error.message}`);
  } else if (error instanceof ProviderError) {
    // Handle provider errors
    showError(`Provider error: ${error.message}`);
  }
}

Migration from v2​

The v2 adapter provides backward compatibility:

// src/lib/server/ai/agent-v2-adapter.ts
export async function executeV2Adapter(
  prompt: string,
  options: V2Options
): Promise<V2Response> {
  const v3Agent = new MolosAgent(options);

  const stream = await v3Agent.stream(prompt, {
    onProgress: (event) => {
      // Convert v3 events to v2 format
      sendV2Progress(convertToV2Progress(event));
    }
  });

  return convertToV2Response(await stream.result);
}

Getting Started​

Basic Usage​

import { MolosAgent } from '$lib/server/ai/agent/v3/core/molos-agent';

const agent = new MolosAgent({
  provider: 'openai',
  model: 'gpt-4-turbo',
  apiKey: process.env.OPENAI_API_KEY
});

const result = await agent.execute('Analyze the latest sales data');
console.log(result.content);

Streaming Responses​

const agent = new MolosAgent(options);

const stream = await agent.stream('Generate a report', {
  onProgress: (event) => {
    console.log(`Step ${event.stepNumber}: ${event.message}`);
  }
});

for await (const chunk of stream) {
  process.stdout.write(chunk);
}

Using Modules​

const agent = new MolosAgent({
  provider: 'anthropic',
  model: 'claude-3-opus',
  modules: ['sales', 'inventory', 'analytics']
});

// Agent automatically discovers tools from modules
const result = await agent.execute('Check inventory status');

Provider Configuration​

OpenAI​

const agent = new MolosAgent({
  provider: 'openai',
  model: 'gpt-4-turbo-preview',
  apiKey: process.env.OPENAI_API_KEY,
  baseUrl: 'https://api.openai.com/v1'
});

Anthropic​

const agent = new MolosAgent({
  provider: 'anthropic',
  model: 'claude-3-opus-20240229',
  apiKey: process.env.ANTHROPIC_API_KEY
});

Z.AI (OpenAI-compatible)​

const agent = new MolosAgent({
  provider: 'openai-compatible',
  model: 'zai-v3',
  apiKey: process.env.ZAI_API_KEY,
  baseUrl: 'https://api.z.ai/v1',
  useCompatibilityMode: true
});

Performance Improvements​

  • Faster streaming: Optimized message chunking reduces latency
  • Parallel execution: Multiple tools can execute concurrently
  • Smart caching: Repeated tool calls are cached automatically
  • Efficient memory: Improved memory management for long conversations

View on GitHub

What's Next​

Future enhancements include:

  • Multi-agent collaboration capabilities
  • Improved tool discovery and suggestion
  • Custom tool registration API
  • Agent persistence and conversation history

Learn about the technical implementation in Devlog Feb 8: AI Agent Improvements.