Introduction

AICP (AI Inference Control Plane) is a provider-agnostic gateway that sits between your application and AI inference providers. Instead of integrating each provider separately, you make a single API call to AICP and it handles routing, fallback, cost tracking, and policy enforcement.

Quickstart

Make your first request in 5 minutes.

JavaScript SDK

Install, authenticate, and stream your first completion.

Python SDK

Sync and async clients for Node-free environments.

Concepts

How routing, tiers, and projects actually work.

Reference

Pricing and the full provider and model catalog.

FAQ

Short answers to the questions we hear most.

How it works

Your appAICP SDKAICP Gateway

OpenAI

Anthropic

Google Gemini

…and every other connected provider

Your application always talks to one endpoint. The gateway selects the best provider based on your routing configuration, checks policies, meters usage, and returns the response transparently. Which providers are live is a running catalog: some connected platform-wide by your admin, others connected by you directly with your own API key (BYOK), not a fixed number.

Key concepts

Gateway

Central API endpoint. Accepts OpenAI-compatible requests.

Providers

AI inference providers your admin has connected platform-wide, or that you've connected yourself with your own key (BYOK). Your own key always takes priority for that provider. See Providers & Models for the full catalog.

Routing

Rules that pick the provider for each request: cost, latency, quality, or balanced.

Projects

Isolated workspaces with their own routing config, budgets, and API keys (Cloud plan and above).

Plans

Community (free, self-hosted), Cloud (hosted, usage-based), or Enterprise (custom). See Pricing.

API Keys

Bearer tokens (aicp-…) used to authenticate SDK and direct API calls.