ZeroTrace Companion
Troubleshooting
Common issues — connection, AI, performance — with the fastest path to resolution.
This page collects the issues that come up most often. For installation-specific issues, see installation.
Connection issues
My device doesn't appear in the picker
Common causes, in rough order of likelihood:
| Cause | Fix |
|---|---|
| Charge-only USB cable | Replace with a data cable |
| Device not powered | Check power LED; unplug, wait 5s, replug |
| USB hub issues | Connect directly to a USB port |
| Driver missing (Windows / older macOS) | Install CP210x or CH340 driver |
| Permission denied (Linux) | Add user to dialout group, log out and back in |
| Port busy | Close other application using the port |
After any fix, click the picker's refresh (or Ctrl+R).
Picker shows "Unknown device"
The device responded but Companion did not match its handshake. Possibilities:
- Custom firmware — define a custom device profile.
- Different ZeroTrace product than expected — ensure firmware is up to date.
- Genuinely non-ZeroTrace device — Companion is for ZeroTrace hardware.
You can still connect via the terminal for raw access.
Connection drops immediately after connecting
| Cause | Fix |
|---|---|
| Cable is loose | Reseat; check for bent connector |
| Device firmware crashed | Hard-reset (hold reset 10s) |
| Conflicting application | Other app may be probing the port — close it |
| OS suspend / resume | Reboot if other fixes don't help |
Auto-detection takes forever
Each port probe has a short timeout. Many ports = longer total scan. If you have many virtual COM ports (Bluetooth virtual devices, software-defined ones), the scan can take a few seconds. Accept it, or reduce the number of virtual COM ports.
AI assistant issues
"AI assistant unavailable"
Most common: the model server isn't running.
# Verify Ollama is running
curl http://localhost:11434/api/tags
Should return JSON (even empty {"models":[]}).
If not, start your runner. On macOS the Ollama menu-bar icon shows status. On Linux/Windows, run ollama serve.
"Model not found"
The model name in Companion's settings doesn't match a pulled model. Check ollama list and confirm the exact name.
Responses take forever
Model is too big for your RAM, or you're CPU-only on a model that needs GPU acceleration. Switch to a smaller model — qwen2.5:3b or similar. See models.
Responses are nonsense
Model is too small for the question, or the model has an issue. Try:
- A larger model.
- A different model family.
- Clearing the conversation (
Ctrl+L) — sometimes the context has gone off the rails.
Tool calls fail
Some smaller models do not support structured tool calls. Use a tool-calling-capable model — qwen2.5:7b is a safe baseline.
AI privacy concern
If you're worried about what the assistant has seen: the assistant's data stays local with default settings. See AI privacy. Disable AI entirely if your posture requires it.
AirLeak issues
Live view shows zero events
| Cause | Fix |
|---|---|
| AirLeak in wrong mode | Check the mode picker; switch to mixed |
| AirLeak hardware failure | Check device LEDs; restart device |
| Firmware mismatch | Update AirLeak firmware to latest |
| Companion not parsing | Check status bar parse-error counter; high = firmware mismatch |
Sessions consume too much disk
Sessions can grow large in busy environments. Mitigations:
- Capture shorter sessions.
- Capture during specific event windows rather than continuously.
- Periodically delete old sessions you no longer need.
- Compress old sessions externally (
gzipreduces them 5-10x).
Library merging seems wrong
The library is conservative — it errs toward "two entries that might be the same" over "one entry that's wrong." If you see two entries that should be merged:
- Click both → Merge in the per-device detail.
If you see one entry that should be split, contact support — manual split is not currently exposed in the UI.
Alerts firing constantly
Tune the alert rules. Common patterns:
- Lower the rule's severity if you're seeing too many.
- Add an exception for the specific noise device.
- Disable the rule entirely if it doesn't fit your environment.
Performance issues
Companion uses a lot of RAM
Most RAM usage comes from:
- Active AirLeak session — millions of events accumulate.
- Open library — paged from disk but caches recent entries.
- AI assistant model — multi-gigabyte memory use.
Mitigations:
- Stop unused sessions.
- Restart Companion after long-running multi-hour sessions.
- Use a smaller AI model.
Companion freezes briefly
Usually correlates with:
- A large session export.
- A library re-index.
- A model loading into memory.
Wait it out — these complete in seconds to minutes depending on size.
Charts in the live view are choppy
High event rates in the AirLeak workspace can saturate the rendering. Mitigations:
- Apply filters to reduce visible event count.
- Pause live updates (
Space) when reviewing. - Use the static devices view instead of the live view for heavy review.
Export and data issues
Exported PDF has broken layout
Long unbroken strings (full JWT tokens, very long MAC lists) overflow page width. Mitigations:
- Edit findings to break long content.
- Use Markdown export instead.
CSV export is missing data
CSV is a flat format; deeply nested results get lossy. Use JSON for full fidelity.
Cannot save export
Most common cause: the chosen path is not writable. Try a known-writable location (Documents, Desktop). On macOS, ensure Companion has been granted file-system access.
When to contact support
You've tried the obvious:
- Latest Companion version installed.
- System clock accurate.
- Permissions granted.
- Re-scanned ports / re-installed driver / replaced cable.
…and the issue persists. Contact support with:
- Your operating system and version.
- Companion version (visible in About dialog).
- Connected device model and firmware version.
- Exact error message or behaviour.
- Reproduction steps.
The more specific the reproduction, the faster the response.
Many issues that look like Companion bugs are actually device-firmware or OS-driver issues. Reproducing the issue with a different device or different machine is one of the most useful diagnostic steps.