Ai Sdk Agents

Multi-agent orchestration with AI SDK v5 - handoffs, routing, and coordination for any AI provider.

Build sophisticated multi-agent systems with automatic handoffs, intelligent routing, and seamless coordination across Ollama (FREE), OpenAI, Anthropic, Google, and other AI providers.

💰 NEW: Use Ollama for zero-cost local AI agents - eliminate $30-200/month in API fees!

🎯 What This Plugin Does

Transform complex workflows into multi-agent systems where specialized agents:

• Hand off tasks to each other automatically
• Route requests to the best-suited agent
• Coordinate complex workflows across multiple LLMs
• Specialize in specific domains or tasks
• Work together to solve problems beyond single-agent capabilities

🚀 Quick Start

Installation

# Install the plugin
/plugin install ai-sdk-agents@claude-code-plugins-plus

# Install dependencies in your project
npm install @ai-sdk-tools/agents ai zod

Your First Multi-Agent System

/ai-agents-setup

# Creates:
# - agents/
#   ├── coordinator.ts      # Routes requests
#   ├── researcher.ts       # Gathers information
#   ├── coder.ts           # Writes code
#   └── reviewer.ts        # Reviews output
# - index.ts              # Orchestration setup
# - .env.example          # API keys template

⚠️ Rate Limits & LLM Provider Constraints

Multi-agent systems multiply API costs - 5 agents × $0.03/request = $0.15 per workflow. Use Ollama (FREE) to eliminate costs entirely.

Quick Comparison: Paid APIs vs Ollama (FREE)

Annual Savings: $360-3,600 by using Ollama for multi-agent orchestration.

Rate Limits by Provider

OpenAI (Paid)

• GPT-4: 10,000 requests/day (Tier 1), 500 RPM
• GPT-4 Turbo: 30,000 requests/day (Tier 2), 3,000 RPM
• Registration: ✅ Email + payment required
• Cost: $30-60/1M tokens

Multi-Agent Impact: 5 agents × 500 RPM limit = effective 100 RPM per agent

Anthropic (Paid)

• Claude Sonnet: 50,000 requests/day (Tier 1), 1,000 RPM
• Claude Opus: 50,000 requests/day (Tier 1), 1,000 RPM
• Registration: ✅ Email + payment required
• Cost: $15-75/1M tokens

Multi-Agent Impact: 5 agents × 1,000 RPM limit = effective 200 RPM per agent

Google Gemini (Paid/Free Tier)

• Gemini 1.5 Flash (Free): 15 RPM, 1M tokens/day
• Gemini 1.5 Pro (Free): 2 RPM, 32K tokens/day
• Gemini (Paid): 1,000 RPM, unlimited tokens
• Registration: ✅ Google account required
• Cost: Free tier available, $0.35-1.05/1M tokens (paid)

Multi-Agent Impact: Free tier 15 RPM = 3 RPM per agent (5 agents) → Very restrictive

Ollama (FREE - Self-Hosted)

• Requests: ∞ Unlimited (hardware-limited only)
• Models: Llama 3.2, Mistral, CodeLlama, etc.
• Registration: ❌ Not required
• Cost: $0 (one-time hardware: $0-600)

Multi-Agent Impact: No API limits! Only limited by CPU/RAM. See ollama-local-ai plugin for full hardware constraints documentation.

Multi-Agent Coordination Strategies

Strategy 1: Shared Ollama Instance (RECOMMENDED - FREE)

// All agents share one local Ollama instance
import { ollama } from 'ollama-ai-provider';

const agents = {
  coordinator: createAgent({
    model: ollama('llama3.2'),  // FREE
    name: 'coordinator'
  }),
  researcher: createAgent({
    model: ollama('llama3.2'),  // FREE
    name: 'researcher'
  }),
  coder: createAgent({
    model: ollama('codellama'),  // FREE
    name: 'coder'
  }),
  reviewer: createAgent({
    model: ollama('llama3.2'),  // FREE
…

Hardware Requirements:

• 4 agents × Llama 3.2 7B: 32GB RAM minimum
• Concurrent requests: Limited by CPU cores
• See: ollama-local-ai plugin for detailed hardware sizing

Annual Cost: $0 (vs $360-3,600 for cloud APIs)

Strategy 2: Hybrid (Free for Development, Paid for Production)

const MODEL_CONFIG = {
  development: {
    coordinator: ollama('llama3.2'),      // FREE
    researcher: ollama('llama3.2'),       // FREE
    coder: ollama('codellama'),           // FREE
    reviewer: ollama('llama3.2')          // FREE
  },
  production: {
    coordinator: anthropic('claude-sonnet'),  // $15/1M tokens
    researcher: anthropic('claude-sonnet'),   // $15/1M tokens
    coder: anthropic('claude-sonnet'),        // $15/1M tokens
    reviewer: anthropic('claude-sonnet')      // $15/1M tokens
  }
};

const models = MODEL_CONFIG[process.env.NODE_ENV || 'development'];