Repository navigation
Accessibility error #987
Replies: 1 comment
|
Two separate things. Whisper offline. The cache layout is the problem, not the network check. Jarvis first calls faster-whisper's Simplest fix: skip the cache and point the setting at a folder. Put "whisper_model": "/Users/<you>/whisper/tiny",
"whisper_backend": "faster-whisper"Jarvis checks the four files and loads from that path with If you would rather keep the Hub cache, make it look like one: cd ~/.cache/huggingface/hub/models--Systran--faster-whisper-tiny
mkdir -p refs
printf d90ca5fe260221311c53c58e660288d3deb8d356 > refs/main
mv snapshots/main snapshots/d90ca5fe260221311c53c58e660288d3deb8d356
# copy vocabulary.txt into that snapshot folder as well
Accessibility. The "This process is not trusted" line is printed by macOS when pynput's Quartz listener creates its event tap for the dictation hotkey. The grant is per executable and per launching app. Running from source, the process is |
Uh oh!
There was an error while loading. Please reload this page.
I’m running into two main issues when starting the daemon on macOS (Intel):
Accessibility warning loop: Both Jarvis and Terminal are enabled in System Settings -> Privacy & Security -> Accessibility, but the console still outputs This process is not trusted! Input event monitoring will not be possible. It seems like a false-positive check.
Whisper offline loading: Because of my network restrictions, auto-downloading faster-whisper-tiny times out (ReadTimeout). I manually placed model.bin, config.json, and tokenizer.json into ~/.cache/huggingface/hub/models--Systran--faster-whisper-tiny/snapshots/main, but the script still tries to connect to Hugging Face and fails.
Logs:
Plaintext
00:14:03 ?? Activity log ready
00:14:04 Starting event loop...
00:14:06 ?? Jarvis daemon started
00:14:06 ?? Initializing daemon components...
00:14:06 ? Daemon started
00:14:06 ?? Using chat model: qwen3.5:0.8b
00:14:06 ?? Using whisper model: tiny
00:14:06 ?? Discovering MCP tools from 3 server(s)...
00:14:11 ? youtube-transcript: 1 tools available
00:14:11 ? macos: 2 tools available
00:14:11 ? chrome-devtools: Connection closed
00:14:11 ?? Initializing dialogue memory...
00:14:11 ? Dialogue memory initialized
00:14:11 ?? Location: nuh uh
00:14:11 ?? Initializing TTS engine (piper)...
00:14:12 ? TTS engine started
00:14:12 ?? Preparing speech recognition in the background...
00:14:12 ?? Warming up models...
00:14:13 ??? Dictation enabled (hold Ctrl + Option to dictate)
00:14:13 This process is not trusted! Input event monitoring will not be possible until it is added to accessibility clients. (!!!!!!!!?)
00:14:14 ?? Checking Whisper model files (first run may download a large model)...
All reactions