Support · InferWatch
InferWatch Support
Getting started
InferWatch monitors your self-hosted LLM inference servers from your iPhone.
- Tap Add demo rigs to explore with sample data, or tap + to add your own server.
- Enter the server's Base URL and pick the type:
- vLLM: the host:port that serves vLLM's Prometheus
/metrics(v0 and v1 metric names both work). - Ollama: your Ollama host (default
:11434). - LiteLLM (gateway): your LiteLLM proxy host.
- Prometheus: a
/metricsendpoint that exposes vLLM-format metric names. Other backends (llama.cpp, TGI, SGLang) use different names and are not mapped yet.
- vLLM: the host:port that serves vLLM's Prometheus
- An optional bearer token is stored in your device Keychain. It is sent only over HTTPS; InferWatch never sends it to a plain
http://address.
Frequently asked
My server won't connect
Make sure the app can route to it. A LAN IP (e.g. 192.168.1.50:8000) needs the local-network permission, so allow it when prompted. A remote box is easiest over a VPN like Tailscale. Confirm the metrics endpoint loads in a browser first.
My server uses a self-signed certificate
InferWatch can't accept a self-signed certificate: it uses the standard iOS certificate check and has no option to approve or pin one. Use a certificate iOS already trusts (for example a Tailscale HTTPS certificate or Let's Encrypt), install your own certificate authority on the iPhone and turn it on under iOS Settings > General > About > Certificate Trust Settings, or use plain http:// on your home network or VPN. Over plain http InferWatch does not send your token.
How do I change a rig's token or address?
InferWatch 1.0 has no edit screen. On the Servers tab, touch and hold the rig, tap Remove (this also deletes its token from the Keychain), then add it again with the new token or address.
No token metrics from LiteLLM
Enable LiteLLM's prometheus callback so /metrics is served. Without it, InferWatch shows gateway health (up / DB / version) only.
Is my data private?
Yes. InferWatch has no backend and collects nothing. It talks only to the servers you add. See the Privacy Policy.
What's in Pro?
Unlimited rigs and the fleet view, breach alerts, history charts, and the Home Screen widget (small and medium). History charts show recent tok/s and KV-cache readings for vLLM and Prometheus rigs; they are kept in memory while InferWatch runs and start over when the app restarts. Pro is a one-time $14.99 purchase, no subscription. Free covers one real server with the live dashboard, and demo rigs never use up that slot.
How do I restore Pro on a new iPhone?
Sign in with the Apple Account you bought with, open the Pro screen (Settings > Unlock InferWatch Pro, or any lock icon) and tap Restore Purchases there. The Pro screen tells you if no previous purchase was found. When Pro is active, Settings shows InferWatch Pro as Unlocked. If nothing changes, wait a minute with a working internet connection and try again.
What happens if I get a refund?
Refunds are handled by Apple at reportaproblem.apple.com; we cannot issue them. After a refund your saved rigs and settings stay on your iPhone, and Pro features lock again.
When do alerts fire?
With Pro and notifications on, InferWatch sends a local notification when a rig crosses your KV-cache ceiling (default 90%), its waiting queue reaches your limit (default 20), it stalls at 0 tok/s while serving requests, or it stops answering. You get one notification per incident and a recovery notice. In the background, iOS decides when the app can check (no more often than every 15 minutes), so treat it as best effort, not real-time paging.
Contact
Anything else: [email protected].