InferWatch · FAQ
Questions, answered.
Straight answers about InferWatch 1.0. Still stuck? See support or write to [email protected].
Connection
InferWatch says my server is unreachable. What do I check first?
Make sure the Base URL starts with http:// or https:// and includes the port, for example http://100.64.0.5:8000. Pick the right type: vLLM and Prometheus read /metrics, Ollama reads /api/tags, LiteLLM reads /health/readiness. Open that address in Safari on the same iPhone; if Safari can't load it, InferWatch can't either. For a home-network address, allow Local Network access (iOS Settings, Privacy & Security, Local Network). Away from home, connect through your VPN. Each request gets about 10 seconds.
The rig shows HTTP 401 or HTTP 403.
The server is up but wants a token. Add your bearer token and make sure the Base URL starts with https://: InferWatch sends the token only over HTTPS. There is no edit screen in 1.0, so to change a token, touch and hold the rig on the Servers tab, tap Remove, and add it again.
My server redirects and InferWatch stops.
InferWatch follows a redirect only to the same host and port (an upgrade from http to https on the same host is fine). Enter the final address as the Base URL.
Do I need to open my server to the internet?
No. Reach it the way you already do: a LAN address on the same Wi-Fi, or a VPN like Tailscale.
HTTPS and tokens
My server uses a self-signed certificate.
InferWatch uses the standard iOS certificate check and can't accept a self-signed certificate. Use a certificate iOS already trusts (Tailscale HTTPS or Let's Encrypt), install and trust your own certificate authority on the iPhone, or use plain http:// on your home network or VPN (InferWatch then sends no token).
Where is my bearer token stored?
In the iOS Keychain on this iPhone, one entry per rig. It is sent only to the server you entered, only over HTTPS. When you remove a rig, its token is deleted.
Alerts, widget and history
Why are breach alerts late?
While InferWatch is open it checks every 2, 3 or 5 seconds. When it is closed, iOS decides when InferWatch may check: at most every 15 minutes, and often less, depending on battery, usage and Low Power Mode. Treat background alerts as best effort, not real-time paging. Keep Background App Refresh on and avoid swiping InferWatch away.
I get no notifications at all.
Alerts are part of Pro. On the Alerts tab tap Enable notifications and allow them, then check that breach and recovery alerts are on in Settings. You get one notification per incident and one when the rig recovers. Defaults: KV-cache 90 percent, queue 20 waiting.
How do I add the widget?
Touch and hold the Home Screen, tap Edit, then Add Widget, and pick InferWatch (small or medium). The widget is part of Pro. It updates when InferWatch is open and when iOS allows a background check.
History charts are empty after I reopen the app.
History charts keep the most recent readings while InferWatch runs and start over when the app restarts. They show on vLLM-type rigs and need Pro.
Purchases and privacy
What does Pro cost?
A one-time purchase of $14.99 (the App Store shows your local price), never a subscription. Free covers one real rig with the live dashboard; demo rigs never use that slot.
I bought Pro but it shows as locked.
Sign in with the Apple Account you bought with, open the Pro screen (any lock icon) and tap Restore Purchases. If nothing happens, wait a minute with a working connection and try again.
What happens after a refund?
Refunds are handled by Apple at reportaproblem.apple.com. Your saved rigs and settings stay on your iPhone, and Pro features lock until you buy or restore Pro.
Do you collect my data?
No. There is no account and no server of ours; InferWatch talks only to the rigs you add. Remove a rig to delete its token. Remove your rigs before deleting the app, because iOS can keep Keychain items after an app is deleted.
Coming soon
InferWatch is heading to the App Store.
Drop a line and I'll tell you the day it lands.
InferWatch is an independent app and is not affiliated with or endorsed by the vLLM, Ollama, or LiteLLM projects. Their names are used only to describe compatibility.