JARVIS UltraManualv1.31

Troubleshooting

Install and service problems

Find the message you see. Lines from the installer that stop it start with Stopped:. For the service, uv run scripts/service.py status lists each problem on a problem: line.

The browser says "JARVIS is not answering"

The server is not running, or is still starting. Run:

uv run scripts/service.py status

If it says Status: not running, run uv run scripts/service.py restart. If that fails, the last lines of logs/service.log say why.

"~/jarvis-ultra exists and is not a JARVIS Ultra install. Move it, or set JARVIS_DIR."

The install line found a folder of that name that it did not make, and left it alone. Rename that folder, or install somewhere else: curl -fsSL <your line's address> | JARVIS_DIR=~/apps/jarvis bash (on Windows, $env:JARVIS_DIR = 'D:\jarvis' before the line).

"1.32 did not install: its installer stopped with exit code 1 (…) 1.31 is back in place."

An update replaced the files but the installer failed on the new version, so the update put the old files back and started the old version again. The lines above it say why (often no internet for uv sync). Fix that and run the install line again. logs/update.log in the JARVIS Ultra folder records each try.

"This copy is a git checkout: update it with git pull, then run the installer again."

scripts/update.py only updates an install made from a release. A folder you cloned with git updates with git pull, then ./install.sh.

"The web app is not built yet. Run pnpm build in web/ and reload."

A release comes with the pages built, so this shows only after you changed the web app and its build failed (pnpm build empties web/dist first). Run ./install.sh (or .\install.ps1) again; its step 2 builds them and shows any error.

"Port 8830 is already in use by pid N (…)"

Another program, or a second copy of JARVIS Ultra, holds the port. The message names the process and how to stop it (kill N on macOS and Linux, taskkill /F /PID N on Windows). Stop it, then run the install or uv run scripts/service.py restart again. On Linux, a process of another user shows as pid 0; sudo ss -ltnp names it.

If you started the dashboard by hand with cd api && uv run python -m app while the service runs, it says jarvis-ultra: 127.0.0.1:8830 is already in use. Stop the other server or pass another port. The service is already serving the dashboard; you do not need the second copy.

"No provider credential yet"

Hermes has no model provider to talk to, so the installer stops before creating the team and ends with exit code 2. Sign in from your terminal:

hermes model

Then run ./install.sh again. scripts/install-hermes.sh --check repeats only the Hermes checks. You can also add an API key in the dashboard under Settings > Models. Some messages from the installer call this "Settings > Providers": it is the Providers section of Settings > Models.

"No default model is set"

Hermes has a key but no model for the Orchestrator. Run hermes model, or set the pair yourself, for example:

hermes config set model.default claude-sonnet-5-5
hermes config set model.provider anthropic

The installer's own hint shows a single hermes config set model <provider/model-id>; use the two lines above instead. Then run ./install.sh again.

"unexpected answer from the default agent"

Hermes is installed but its test turn did not answer. The lines under the message show what it said, usually a refused key, no credit, or a model the provider does not offer. Fix that (see When an agent does not answer), then run scripts/install-hermes.sh --check. The installer may report this as "Hermes did not install"; the install itself is fine.

The installer keeps ending with exit code 2

Read the Left to do: list. The usual item is "Tell the agents who you are: copy scripts/team/owner.example.yaml to data/owner.yaml…". Filling in Meet your team in the dashboard, or Settings > General > Owner, does exactly that; after it, the installer ends with 0.

"The service is installed but http://127.0.0.1:8830/api/ping does not answer."

The service started but the dashboard did not come up within 30 seconds. Run uv run scripts/service.py status and read the last lines of logs/service.log.

Linux: "The user service manager did not answer"

Run the command as your own user from a login shell (a terminal or ssh), not through sudo. If it keeps failing, switch linger on once:

sudo loginctl enable-linger $USER

Linux: linger is off

status notes "linger is off, so after a reboot JARVIS Ultra starts only once you logs in". Run the loginctl line above once.

Linux: "This Linux has no systemd"

The service needs systemd. Install with ./install.sh --no-service and start the dashboard yourself with cd api && uv run python -m app. On WSL2, switch systemd on instead (Install on Linux).

Linux: libatomic.so.1: cannot open shared object file

Node.js, which the installer fetches only to build a web app you changed, needs a small library. Install it, then run ./install.sh again:

sudo apt-get install libatomic1

On Fedora: sudo dnf install libatomic. The installer does this itself when it may use sudo.

Voice did not install

On macOS and Linux, the voice installer in this version stops at once with VENV_PY: unbound variable, and ./install.sh --voice lists "Voice did not install" under Left to do:. Until that is fixed, run the voice installer with that variable set, from the JARVIS Ultra folder:

VENV_PY="$PWD/voice/.venv/bin/python" bash scripts/install-voice.sh

On Linux the voice installer makes its own Python 3.12 with uv, so no python3-venv package is needed; Piper runs on x86_64 and arm64. uv run scripts/service.py status --voice shows the voice service. On Windows, Piper is untested; the Browser voice works without it.

Restore says "A folder already exists at …/profiles/name. Move it away first."

This happens when one of the five original specialists was retired and the installer ran again: it made a fresh copy of that agent. To bring back the retired one, move the fresh folder out of profiles/ (for example to your Desktop), then press Restore again.

Windows

  • "Running scripts is disabled on this system": run powershell -ExecutionPolicy Bypass -File .\install.ps1.
  • hermes is not recognized right after the install: the PATH changed for new windows only. Open a new PowerShell window.
  • \r: command not found from a script: the clone has Windows line endings. Delete the folder and clone again.
  • "Task Scheduler refused the task": group policy blocks scheduled tasks. The installer put a shortcut in your Startup folder instead. It starts the dashboard at logon but not after a crash; uv run scripts/service.py restart brings it back by hand.
  • "Port 8830 is already in use by pid N (python.exe)": another copy runs. End it with the taskkill /F /PID N the message gives, then run restart.