Troubleshooting¶
Fixes for the problems you are most likely to hit with Tabby Coding Assistant. Each entry gives the symptom, the usual cause, and what to do.
Completion requests are refused¶
Cause: no completion model is attached. The image ships without model
weights, so /v1/completions refuses requests until you add one.
Fix: attach a model as described in Attach a model, then restart Tabby.
Nothing answers on port 8080¶
Cause: first boot has not finished, or Tabby or the embedding stub failed
to start. tabby.service does not start without sushi-tabby-embed.service
and sushi-tabby-firstboot.service.
Fix: check all three units and their logs:
systemctl status sushi-tabby-firstboot sushi-tabby-embed tabby --no-pager
sudo journalctl -b -u sushi-tabby-firstboot -u sushi-tabby-embed -u tabby --no-pager
If you edited /var/lib/tabby/config.toml, look for a configuration error in
the tabby log, fix it, and restart Tabby.
Tabby stops or restarts after I attach a local model¶
Cause: the model needs more memory or disk than the instance has, or the download cannot reach the internet.
Fix: check the log for out-of-memory or download errors. Resize to a larger instance type and volume, allow outbound HTTPS, then restart Tabby.
The self-test fails after I attach a model¶
Cause: smoke.sh and the walkthrough expect the default configuration:
no completion model, and the embedding stub on port 18080.
Fix: this is expected. Check Tabby with /v1/health and a real request
instead. See Check the installation.
I cannot reach Tabby from my computer¶
Cause: by design, Tabby listens on 127.0.0.1 only.
Fix: use Session Manager port forwarding, as in Getting started. Keep the port-forwarding session open while you work.
Permission denied when I read the configuration¶
Cause: /var/lib/tabby belongs to the tabby user with mode 0700.
Fix: use sudo, for example sudo cat /var/lib/tabby/config.toml.
Last updated October 13, 2026 · Suggest an improvement