Troubleshooting

The short list of things that actually go wrong, and what fixes them. ScyllaChat is in beta; when something here does not cover you, report it and it usually gets fixed fast.

Test API fails

• Check Model API is the provider's base URL, for example https://api.openai.com, not a full path with /v1/chat/completions. The app adds the right path itself.
• Check the key is current and has credit. Providers reject expired or unfunded keys with the same error shape as wrong ones.
• Check API Format matches the provider: Claude keys need Claude (Anthropic), OpenAI-compatible services need OpenAI-compatible.

The model list is empty

Press Test API first; the list fills from the provider's models endpoint. Some OpenAI-compatible services do not expose one, in which case type the model name in manually and carry on.

Replies stop mid-sentence

Raise Max Response Tokens in the API tab. If the provider enforces a smaller ceiling, the reply stops at theirs, not yours.

The dot next to the avatars is red

That dot is the app's own local server, which saves your data and passes requests on to your provider. Tap it: it says what is wrong, can check again, and opens the API settings. Red usually means the server process stopped; restarting the app brings it back.

"Error during initialization"

Refresh once. If it persists, check for an update in the Sync tab; initialization failures after an update are treated as release-blocking bugs, so a fix is usually already out.

Android refuses the APK

Android blocks installs from outside its store by default. Allow "install unknown apps" for your browser or file manager, then open the APK again. Updates installed from inside the app do not need this dance a second time.

Paired devices do not sync

• Both devices must be online at the same time; the relay passes ciphertext live and stores nothing.
• Check both devices show the same sync group in the Sync tab. If not, re-join with a fresh invite.
• Press Sync now and watch the log at the bottom of the tab; it names anything that fails.

Streaming acts up

A few OpenAI-compatible proxies stream badly or not at all. Toggle Enable streaming responses off in the chat's ⋯ menu and the same request arrives in one piece instead.

Where the logs are

The browser console (F12 on desktop) carries the app's own logging, including prompt construction and sync activity. When reporting a bug, a console screenshot shortens the round trip enormously.