Start here: private AI for any size, kept current.
Magic Sites

// MAGIC_AI_GATEWAY · BYOK_ROUTING

Your keys. Our guardrails. Better routing for the providers you already use.

Bring your Anthropic, OpenAI, Gemini, or Cohere key. Route every call through Magic AI Gateway for automatic failover, real-time observability, guardrails, team-key management, and usage analytics. Your provider tokens stay billed to your provider account. We charge only for the management layer.

// WHAT_YOU_GET

Six layers on top of your provider keys.

BYOK - Bring your own keys

Anthropic Claude, OpenAI GPT, Google Gemini, Cohere - route any provider through our Gateway. Your tokens stay billed to your provider's account. We charge only for the management layer.

Failover routing

If your primary model degrades or rate-limits, the Gateway falls back to your declared backup automatically. Zero downtime to your application.

Guardrails

Per-key cost caps, prompt-injection detection, output filtering, PII redaction, and content-policy enforcement. Configurable per-team, per-environment.

Team keys + access control

Issue scoped sub-keys to teammates with usage limits, model allowlists, and revocation. WorkOS SSO + audit log.

Observability + analytics

Every call logged with latency, cost, model, route, and outcome. Dashboard + webhook export. Spot regressions, cost spikes, and usage patterns in real time.

OpenAI-compatible router

Drop-in replacement at the SDK level. Same client code, smarter routing. Switch between Inference (curated open-weights) and Gateway (your provider keys) by changing the model parameter.

// WHO_USES_IT

Five real workloads.

  • Mid-market ops team with existing OpenAI Enterprise contract - wants observability + cost guardrails
  • Agency reselling AI to clients - needs per-client team keys + usage caps + white-label dashboard
  • Compliance-heavy customer - wants PII redaction + audit log + provider failover for SLA
  • Startup A/B-testing models - routes 50/50 between Claude and GPT to compare quality + cost
  • Production app - needs primary-Anthropic with fallback-to-Gemini if Anthropic degrades

// PRICING

You pay your provider. We bill the management layer.

Magic AI Gateway is metered on calls routed, not on tokens consumed (your tokens go to your provider). Volume tiers + team-key bundles in detail by request - for now we're piloting with a small group of customers and iterating on price points. Talk to us with your monthly call volume + provider mix and we'll quote.

// MAGIC_LLM_HUB

Magic AI Gateway is part of the Magic LLM hub. Want curated open-weights at retail rates instead of BYOK routing? See Magic Inference. Need both? Same dashboard handles both paths.