A3M Router - One prompt in. The right model out.

# Your app today — every request goes to GPT-4o
const response = await openai.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Explain quantum entanglement" }]
});

❌ $0.003 per request — no fallback if OpenAI goes down

# One line change — A3M routes to the cheapest capable model
const response = await fetch("http://localhost:8787/v1/chat/completions", {
  body: JSON.stringify({
    model: "auto",  // ← A3M picks the best model
    messages: [{ role: "user", content: "Explain quantum entanglement" }]
  })
});

✅ Routed to: Groq (FREE)

$0.00 per request — automatic fallback chain if Groq fails

🌏 47+ Providers

Chinese Providers (special handling)

Groq (FREE) DeepSeek Kimi Qwen Zhipu Yi Stepfun Moonshot 01.AI Tencent Baidu Alibaba

Global Providers

OpenAI Anthropic Google Mistral Cohere Groq Perplexity AWS Bedrock Azure Replicate HuggingFace

✅ A3M routes to the cheapest healthy provider

Health checks + circuit breakers + intelligent fallback

💰 Save 62% on API Costs

WITHOUT A3M

$0.003

per request (GPT-4o)

$3.41 per 1K requests

WITH A3M

$0.00

47% → Groq (free)

30% → DeepSeek ($0.14/M)

23% → GPT-4o-mini ($0.15/M)

$1.24 per 1K requests

62%

Cost Savings

99.5%

Routing Accuracy

<1ms

Routing Latency

💰 $2,175 saved per 1M requests

At 1000 queries/day: $547 saved yearly

✨ Features

🧠 Adaptive Memory

Learns from your usage. Updates model quality scores with every request. No retraining.

💾 Semantic Cache

Embedding-based lookup. 30%+ hit rate on repeated queries.

🛡️ Budget Enforcement

Per-user/team caps. Alerts at 50%/80%/100%. No surprises.

🔄 Intelligent Failover

Circuit breaker (3 fails → 60s cooldown). Automatic fallback chains.

⚡ Per-Provider Retry

Custom timeout per provider. Exponential backoff. 429 handling.

🎯 12-Signal Routing

Domain, task, structure, action verb, multi-step. Zero ML.

Zero ML. No GPU required. Starts in <100ms.

🔄 Intelligent Failover

[Groq] — Attempting connection... [Groq] — ✗ FAILED — 503 Service Unavailable [Circuit Breaker] — Tripped after 3 failures [DeepSeek] — HEALTHY — Switching... [DeepSeek] — ✅ Response delivered ✓

✅ Your app never knew there was a problem

99.9% uptime — automatic provider health scoring + fallback

📊 Provider Health Monitoring

Latency + error rate → health score → automatic routing

Get started in 10 seconds

npm install -g adaptive-memory-multi-model-router

# Auto-detects your API keys — zero config

npx a3m-router serve

# Now your app: model: "gpt-4o" → model: "auto"

1 line

Code change

0

API keys to manage

<100ms

Startup time

✨ One prompt in. The right model out. ✨

🔀 A3M Router

🌏 47+ Providers

💰 Save 62% on API Costs

✨ Features

🔄 Intelligent Failover