Delamain ghcr.io/vonhex/delamain:latest https://github.com/vonhex/delamain/pkgs/container/delamain bridge bash false https://github.com/vonhex/delamain/issues https://github.com/vonhex/delamain Delamain is a self-hosted voice companion for the drive, modeled on Delamain from Cyberpunk 2077. Open it in any browser, including the one on your car's Android Auto display, press the mic and talk. It answers in a cloned Delamain voice (F5-TTS), searches the web through SearXNG, reads you spoken news bulletins, keeps notes, reminders and long-term memory in your Obsidian vault, and hands navigation to your phone's map app. A corrupted personality core with stutters and a glitched voice can be switched on in Settings. Works with any OpenAI-compatible LLM that supports tool calling (llama.cpp, vLLM, Halogen). F5-TTS model weights and a voice reference clip must be supplied by the user (see GitHub). AI: Productivity: Tools: http://[IP]:[PORT:8888]/ https://raw.githubusercontent.com/vonhex/unraid-templates/main/Delamain.xml https://raw.githubusercontent.com/vonhex/delamain/main/logo.png Requires an OpenAI-compatible LLM endpoint with tool calling. F5-TTS model weights must be downloaded separately (see GitHub README), and a 10-second voice reference clip placed in App Data as delamain_ref_10s.wav. After the first start, set a password from the container console: python auth.py set-password YOURPASSWORD. A GPU is recommended for real-time TTS; CPU-only works but is slower on uncached phrases. 8888 /mnt/user/appdata/delamain/ /mnt/user/appdata/delamain/checkpoints/ http://192.168.1.x:8080/v1/chat/completions true http://[IP]:8888