What is ZeaLLM?
ZeaLLM is a self-hosted AI gateway. Applications call one OpenAI-compatible endpoint with a zea- virtual key. The gateway authenticates the key, enforces budgets, rate limits, model access and guardrails, then routes the request to OpenAI, Azure OpenAI or Anthropic.
The platform has two services that share one Postgres database:
| Component | Role | Default URL |
|---|---|---|
| Gateway | Data plane. Every LLM call goes through here. | http://localhost:8080 |
| Portal | Control plane. Keys, models, budgets, teams, audit. | http://localhost:3000 |
This documentation site runs on http://localhost:3002. In the portal, Documentation in the sidebar (and the /docs path) opens this site in a new tab. The portal's in-app API reference page lists the gateway endpoints with copy-paste examples.
What you can do
- Issue virtual keys through a request → approve → reveal-once workflow
- Route traffic across weighted model deployments with retries and fallbacks
- Enforce hierarchical budgets (key, user, application, team, organization, tag, customer)
- Apply guardrails by bundling Garden filters into a policy and attaching it to a key
- Attribute spend with
x-zeallm-tagsandx-zeallm-customer-id - Review usage, request logs and an audit trail in the portal
Start here
- Architecture — how the pieces fit together
- Local setup — run the stack
- First request — send a chat completion through the gateway
If you already have a revealed key, jump to the SDK snippets or your role guide. New to the portal? Start with Navigating the portal.