When something goes wrong, Jarvis shows a bar at the top of its window with what happened and a button that fixes it. This page lists those messages and a few problems that don't show a bar. If nothing here helps, see "Reporting a problem" at the end.
Messages in the bar#
| Message | What it means | What to do |
|---|---|---|
| "… can't use the microphone because it's turned off in privacy settings" | Windows or macOS blocks microphone access for Jarvis | Open privacy settings (on a Mac: Open System Settings), turn on microphone access (on Windows also for desktop apps; on a Mac for Jarvis), then restart the assistant |
| "Your microphone was disconnected or isn't available" | The chosen microphone is gone | Choose another microphone in Settings › Microphone, or plug it back in. Typing works meanwhile |
| "Speech recognition isn't downloaded yet" | The Whisper model isn't on disk | Open Models and download it, or wait for the download to finish |
| "Voice input isn't working right now" | The voice pipeline couldn't start | Open Settings › Microphone, then the log (Settings › Logging and diagnostics) shows the reason |
| "…'s voice couldn't start, so answers are shown as text" | The chosen voice can't load: not downloaded, or its engine isn't installed | Open Models and click Download on the voice. For Kokoro and Piper voices this also installs the engine. Or pick the Windows voice |
| "Claude needs Claude Code, Anthropic's app, which isn't installed" | Sign in with Claude uses Claude Code | Get Claude ready installs it; or install it from Anthropic's setup page |
| "You're signed out of Claude" | Claude Code's login expired | Sign in and finish in the browser |
| "Claude's usage limit is reached" | Your Claude plan's limit for now | Wait until the time shown, or Use an API key instead |
| "This API key didn't work" | The key is mistyped, revoked, or for another provider | Change key in Settings › AI and Test connection |
| "You're offline, or the AI service can't be reached" | No internet, or the provider is down (for Ollama / LM Studio: the program isn't running) | Check the connection, start Ollama or LM Studio, then Try again |
| "The AI connection isn't set up yet" | No model, key or server address chosen | Open Settings › AI and finish the connection |
| "Today's AI spending limit is reached" | Your daily limit for API providers | Wait until tomorrow, or Change the limit |
| "A model didn't finish downloading" | The download stopped (connection, disk space) | Retry. Downloads resume where they stopped |
Problems without a message#
"Windows protected your PC" when opening the installer. Jarvis isn't code-signed, so SmartScreen warns. Click More info, then Run anyway. To be sure the file is genuine, compare its SHA-256 with the one on the release page (INSTALL.md).
Mac: "Jarvis can't be opened" or "Apple could not verify Jarvis". Jarvis isn't notarized by Apple (that needs a paid developer account). Open System Settings › Privacy & Security, scroll down to the message about Jarvis, click Open Anyway, then Open. Right-clicking the app and choosing Open no longer skips this since macOS 15. To be sure the app is genuine, compare the disk image's SHA-256 with the one on the release page (INSTALL.md).
Mac: "Jarvis is damaged and can't be opened". The download was cut short or changed on the way. Delete it, download it again and compare its SHA-256 with the one on the release page.
Mac: Full control doesn't work, or says macOS hasn't allowed it. Clicking, typing and reading other apps need Accessibility, screenshots need Screen & System Audio Recording, both in System Settings › Privacy & Security. Turn on Jarvis in each (Settings › Tools and permissions has buttons for both pages). Screenshots work after Jarvis is reopened. After an update macOS may ask again, because Jarvis is signed without a developer certificate: turn Jarvis off and on again in those lists.
Mac: Jarvis can't open or type into Terminal. macOS asks once whether Jarvis may control Terminal. If you said no, turn it on in System Settings › Privacy & Security › Automation › Jarvis › Terminal.
Mac: Claude Code isn't found. The setup wizard installs it into ~/.local/bin. If you installed it some
other way, make sure claude works in Terminal; Jarvis also looks in Homebrew's folders.
"The update didn't install: …". The message says what was in the way, often a file another program had open. Restart the computer and try Settings › About › Check for updates again. If it keeps failing, download the new version from the website and install it over the old one: your settings, memories and models stay.
It hears me, but nothing happens. Look at the Activity view. If a tool was denied, its group may be off (Settings › Tools and permissions › What it may do) or the tool is set to deny. If it timed out, nobody answered the approval in time.
It asks before everything after it read a web page. That is the untrusted-content rule: after a web page or a screen, every action that changes something asks first in that request. See SECURITY_MODEL.md. You can turn it off in Settings › Tools and permissions, but leaving it on is safer.
I said "yes" but it still waits. Deleting a file, emailing someone outside your trusted list, shutting down or restarting, and typing into a running program need a click on Allow or the Y key. A spoken yes isn't accepted for those. A spoken "no" always works.
"Desktop control is not allowed in …". Jarvis refuses to click or type in its own windows, consoles, admin prompts, password managers and browser password pages. That is on purpose.
It answers slowly. Speech recognition on the CPU takes a moment, and a bigger Whisper model is slower: Settings › Models shows what suits your computer. Answers depend on the AI provider. Small local models are slower and less reliable with tools; Settings › AI has the options. The Chatterbox voice is slow without an NVIDIA graphics card: about 10 seconds before each answer starts.
The voice sounds robotic. That's the built-in Windows (or Mac) voice, which speaks until the Kokoro voice has downloaded. Check Settings › Models: the Kokoro row should say it's installed.
A Kokoro or Piper voice doesn't install. The voice's engine comes from PyPI, so the PC needs internet
access while it installs. The error in Settings › Models gives pip's reason. From a source install you can also
run python -m pip install -e .[kokoro], or python scripts/setup_piper.py --engine for Piper.
Kokoro starts each answer late. It runs on the processor, and a slow or busy processor takes longer. The first sentence plays as soon as it's ready while the next one is made. Settings › Voice › Kokoro speed doesn't change this; the Piper or Windows voices start sooner.
Reporting a problem#
- Settings › Logging and diagnostics › Save a report (from a source install:
python jarvis_gui.pyw --diagnose). The report masks passwords and keys and hides your user folder and email addresses, but read it before you share it. - Open a bug report on GitHub with what you said or typed, what happened and what you expected, and attach the report.
Security problems go through private reporting instead: see SECURITY.md.