Architecture & deployment

How Coco Router is put together, and how it's run. Coco Router is a fully managed service — there's nothing to install and nothing to host yourself. This page explains the shape of the system so you understand what sits between your application and the model providers.

The request path

Every application talks to one endpoint. Behind that endpoint, a small number of components do the work:

  1. Reverse proxy (TLS). A shared reverse proxy terminates HTTPS at router.cocolevio.com and forwards traffic to the gateway.
  2. Gateway web app. The core service. It authenticates the virtual key, applies budgets and allow-lists, runs Auto-Switch on chat requests, calls the chosen provider, and records the request for usage tracking.
  3. Provider backends. The gateway holds credentials for every provider — OpenAI, Google, the OpenRouter catalog, Amazon Bedrock, and more — and routes each request to the right one. Your application never holds a provider key and never names a provider.

The same service also serves the admin consoles — the Dashboard, Auto-Switch Control Panel, AI Console, and Lead Agent — under their own paths.

Cocolevio's extensions are add-ons

Coco Router is built on a leading open-source LLM gateway, extended by Cocolevio with the Auto-Switch complexity router, the live Control Panel, the usage and cost Dashboard, the AI Console, universal auto-routing keys, and the Lead Agent.

The important architectural principle: the open-source base is never modified. Every Cocolevio feature is layered on top as an add-on — the custom console tabs are injected into the admin UI at serve time, and the routing logic wraps the gateway rather than patching it. That means the underlying platform upgrades cleanly: new base-platform releases can be adopted without reworking the Cocolevio extensions.

Configuration that applies live

The routing configuration — the model pool for each complexity tier and the classifier model — lives in a configuration file the gateway hot-reloads. Changing it in the Control Panel takes effect on the very next request: no restart, no redeploy. See the Control Panel guide for how that works in practice.

Keys, budgets, and per-request spend records are held in the gateway's own data store, which is what powers per-key cost attribution.

Deployment model

Coco Router runs as a managed service operated by Cocolevio:

  • Nothing to install. You get an endpoint and a key; that's the whole integration surface.
  • Credentials are provisioned. There's no public sign-up — workspace and admin credentials come from the Cocolevio team. See Security.
  • Provider onboarding and infrastructure changes — adding a provider key, wiring Amazon Bedrock, scaling — are handled by Cocolevio administrators. See Managing the stack for what admins can do from the consoles, and what needs a platform-level change.