Enterprise AI Gateway
Route, translate, load-balance, and observe multiple LLM upstreams through one entry point
Multi-upstream aggregation · protocol translation · multi-key weighted distribution · load balancing · access control · observability
$ docker pull ghcr.io/gatellm-io/gatellm:latest$ docker run -d \
--name gatellm \
-p 7890:7890 \
-e STORAGE_MODE=sqlite \
-v gatellm-data:/var/lib/protoflux \
ghcr.io/gatellm-io/gatellm:latestZero-config startup: the image enables the Console by default, and on first start it automatically generates an admin password and prints it once. Grab the password to log in to the Console and configure upstreams, models, and access keys. For the full 8 steps (about 5 minutes), see Docker single node walkthrough.
⚠️ This one-minute command is zero-config and does not pass
ENCRYPTION_KEY. The container will start and the Console is accessible, but the first time you save an upstream / issue an access key it will reportencryption_key not set in config. To configure a truly usable upstream, first follow Docker single node walkthrough → Prerequisite to generateENCRYPTION_KEYand start with it set.
If you want a coding agent (Claude Code / Cursor / Codex, etc.) to install it for you, see Agent-assisted install.
This manual is for end users — the people who use the product. Whichever type you are, we recommend first getting Docker single node running, then branching by role:
| You are | What you want to do | Start with |
|---|---|---|
| Integrator / caller | Connect a client application to the gateway and call models through it | Docker single node walkthrough → Endpoints · auth · protocol interop → Client integration and gateway differences → Error codes |
| Administrator / operator | Manage users, access keys, upstreams, models, and routing in the Console, and view statistics and logs | Docker single node walkthrough → Console login and roles → Overview dashboard → Field quick-reference index |
Not sure where to start? Look up your goal below (for the applicable audience, difficulty, and detailed descriptions of each use case, see the Use case overview):
| Goal | Starting point |
|---|---|
| Enhance model capabilities | Enhance model capabilities |
| Reduce call cost | Reduce call cost |
| Achieve high availability / disaster recovery | Achieve high availability |
| Multi-tenant isolation | Multi-tenant isolation |
| Office / Microsoft 365 add-in | Office / Microsoft 365 add-in |
| Meet compliance and audit | Meet compliance and audit requirements |
| Troubleshoot problems | Troubleshoot from symptoms |
| Choosing a request-rewrite approach | Request rewriting and routing: how to choose |
| Production readiness check | Production checklist |
| Script performance and memory | Script performance and memory |
| Access-control design | Access-control design |
http://localhost:7890 by default, and the examples in this manual assume this; if you deploy elsewhere, replace localhost:7890 with your address.http://localhost:7890/console.sk-... in examples is a placeholder for an upstream API key (e.g. an OpenAI / Anthropic key); replace it with your own upstream API key. It is not the gateway access key — the gateway key is written in this manual as <your access key> and is issued by you in the Console.