gateway

The Portkey AI Gateway is an open-source enterprise gateway that routes requests to 1600+ language, vision, audio, and image models through a single, fast API. You can run it locally with Node.js and npm, and then send requests to the gateway's API to access a wide range of models and providers without integrating each provider separately. The gateway handles routing, retries, fallbacks, load balancing, and guardrails to help you manage multi-model deployments reliably. After starting the gateway, you’ll interact with a unified request surface at the gateway URL (for example, http://localhost:8787/v1) and use its console for logging and observability where available. The repository also notes enterprise-ready features like auth and observability through the MCP Gateway, making it suitable for private deployments and managed environments.

Prerequisites:

• Node.js and npm installed on your machine
• Sufficient network access to install and run the gateway

Option A: Run via npx (no local install required)

1. Open a terminal
2. Run:

npx @portkey-ai/gateway

3. The gateway will start and listen on http://localhost:8787/v1 (and the console at http://localhost:8787/public/).

Option B: Install locally (if you want to pin versions or customize)

1. Ensure Node.js and npm are installed
2. Install the package:

npm install -g @portkey-ai/gateway

3. Start the gateway:

gateway  # or npm run start, depending on package.json scripts

4. Access the API at http://localhost:8787/v1 and the console at http://localhost:8787/public/

Notes:

• The gateway is designed to be drop-in compatible with existing MCP tooling. You can manage MCP servers and observability via the gateway’s interfaces.
• If you use a private deployment or enterprise features, consult the enterprise deployment guides referenced in the docs for authentication and topology considerations.

Tips and common issues:

• Default gateway URL is http://localhost:8787/v1; the console lives at http://localhost:8787/public/
• For production deployments, consider enabling authentication and securing the gateway endpoints as described in the MCP Gateway docs.
• The gateway supports retries, fallbacks, load balancing, and guardrails to protect and stabilize multi-model routing.
• If you run into port conflicts, adjust the gateway’s port in the deployment configuration or via environment variables as supported by your deployment method.
• Check the repository’s Quickstart and deployment sections for Docker or cloud deployment options if you don’t want to run locally.