Troubleshooting

Fixes for the daemon not starting, slow or missing completions, download failures, GPU problems and Windows SmartScreen.

Where to look first

  • Status bar (bottom right): shows offline, setup, a model name, or an error. Click it for the status menu.
  • BdebTech AI: Show Logs opens the daemon output inside the editor.
  • Daemon log file: %LOCALAPPDATA%\BdebTech\AI\logs\bdebd.log.

“BdebTech: offline” in the status bar

The daemon did not start or crashed.

  1. Run BdebTech AI: Restart Local AI Daemon.
  2. Check whether another program uses port 41337: netstat -ano | findstr 41337. Change the port in %LOCALAPPDATA%\BdebTech\AI\config.json ("port") and in the setting bdebtech.daemon.url.
  3. If you installed only the extension, make sure bdebd is on PATH or set bdebtech.daemon.path.
  4. Antivirus may quarantine bdebd.exe or llama-server.exe (both are unsigned in the preview). Add an exclusion for the install folder and %LOCALAPPDATA%\BdebTech\AI.

No completions appear

  • Is autocomplete on? The status bar shows (tab off) when disabled; press Ctrl+Alt+Shift+A.
  • The language may be in bdebtech.completions.disabledLanguages (plaintext and markdown are off by default).
  • Another AI extension (Copilot, Cursor Tab, Codeium) may be taking the inline suggestion slot. Disable one.
  • The engine may be loading after idle unload; the first request after 30 minutes takes about a second.

Completions are slow

  • On CPU, make sure no other heavy process is running; llama.cpp uses all physical cores.
  • Reduce bdebtech.completions.contextChars (default 3000) and maxTokens (default 64).
  • Switch to the Starter model if you are on Lite+ with 8 GB RAM.
  • Check the engine variant in the status menu. If you have an NVIDIA GPU and it says cpu, update your GPU driver (CUDA 12.4 runtime needs driver 550+) and restart the engine.

Download failed or stuck

Downloads resume automatically. Run BdebTech AI: Download Starter Model again. If your network blocks Hugging Face or GitHub, download the files on another machine and place them under %LOCALAPPDATA%\BdebTech\AI\models\<model-id>\ and ...\engines\, see Offline mode. Checksums are verified after download; a mismatch deletes the file and retries.

GPU engine crashes on start

The daemon falls back to CPU automatically and logs the reason. Common causes: outdated drivers (CUDA), missing Vulkan runtime on very old GPUs, or not enough VRAM for the chosen model. Force CPU with "engine": "cpu" in config.json if you prefer.

Windows SmartScreen / “Windows protected your PC”

The preview build is not code-signed. Click More info → Run anyway. Compare the SHA-256 on the release page if in doubt.

Another tool gets 401 unauthorized from the daemon

Every request needs the per-install API token (file token in %LOCALAPPDATA%\BdebTech\AI, or bdebd token). Pass it as the API key / Authorization: Bearer header in the other tool. See Local API.

Chat says “engine_unavailable”

No model is installed, or the active model’s files were deleted. Run BdebTech AI: Download Starter Model.

Still stuck

Open an issue with the last 50 lines of bdebd.log at GitHub or message WhatsApp support.