Your first request
This page takes about five minutes and proves the whole thing works. By the end you will have asked anz33 a question and seen which engine answered it.
1. Have at least one engine running
anz33 does not contain an AI. It routes to the ones you already have. If you have none yet, Ollama is the usual starting point — install it, then pull a small model:
ollama pull llama3.2A small model on a normal laptop answers in a couple of seconds. A very large one may take a minute or refuse to load at all, which is a poor first experience — start small.
2. Start anz33
If you installed the app, it is already running: check the menu bar icon (macOS) or the tray icon (Windows). From a terminal:
anz33 serve
anz33 0.4.1 — http://127.0.0.1:3300/v1
dashboard: http://127.0.0.1:3300/anz33dashboardLeave that window open — closing it stops the router. (anz33 install
makes it start on its own at login so you never need this window again.)
3. See what it found
In a second terminal window:
anz33 status
anz33 0.4.1 · plugged in
ollama http://127.0.0.1:11434 3 models
lmstudio not running
llamacpp not runningEngines that are not running are listed as not running rather than hidden — if you expected one to be there, that line is where you find out it is not.
4. Ask it something
curl http://127.0.0.1:3300/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{"model": "auto", "messages": [{"role": "user", "content": "Say hello in five words"}]}'The reply is a block of JSON with the answer inside it. The important part is that you got one.
"model": "auto" is the interesting bit. You did not name a
model or an engine. anz33 picked one — the fastest that could answer. Name a
specific model instead and anz33 finds whichever engine is holding it.
5. Or skip the terminal entirely
Open 127.0.0.1:3300/anz33chat in your browser. It is a plain chat page served by anz33 itself — nothing is loaded from the internet, and you can type a question and get an answer without a single command.
6. See what just happened
Open the dashboard. Your request is on it: which engine took it, how fast it was, how many tokens came back. That page is the answer to "is this thing actually doing anything".