Production AI Toolkit

Protect your AI APIs. One package.

Rate limiting, cost tracking, caching, safety guardrails, and fallback chains for every AI provider. Zero vendor lock-in.

Get Started GitHub

armor.config.ts

import { createArmor } from 'ai-armor'

const armor = createArmor({
  rateLimit: {
    strategy: 'sliding-window',
    rules: [{ key: 'user', limit: 30, window: '1m' }],
  },
  budget: {
    daily: 50,
    onExceeded: 'downgrade-model',
    downgradeMap: { 'gpt-4o': 'gpt-4o-mini' },
  },
  safety: { promptInjection: true, piiDetection: true },
  cache: { enabled: true, ttl: 3600 },
})

Trusted by teams using

OpenAIAnthropicGoogleMistralCohereDeepSeekGroqAWS BedrockAzure+9 more

Rate Limiting

Sliding-window rate limiter with per-user, per-IP, and per-key rules. Shared state across instances via Redis.

Cost Tracking

Real-time cost tracking with 69 model pricing database. Daily, monthly, and per-user budgets with auto-downgrade.

Response Caching

O(1) LRU cache with exact-match strategy. Skip redundant API calls. Custom key functions for control.

Model Routing

Alias models by tier (fast/balanced/best) and resolve at runtime. Switch providers without changing application code.

Safety Guardrails

Block prompt injection, detect PII, enforce token limits, and filter with custom patterns. Fail closed.

Logging & Observability

Structured logs for every request with cost, latency, tokens, and cache status. Stream to any analytics backend.

18+ AI Providers

OpenAI, Anthropic, Google, Mistral, Cohere, DeepSeek, Groq, Together AI, Fireworks, Perplexity, AWS Bedrock, Azure, and more.

Zero Lock-in

Framework-agnostic core. First-class Vercel AI SDK middleware, Express/Hono/Fastify handlers, and Nuxt module.