Local LLM chat interface for Linux.
Usage: Set a keyboard shortcut to launch the script. This opens a prompt dialog, then a terminal for output and further chat (with formatting for markdown and thinking tags, and chat history).
Portable and customizable. Uses a dialog like wofi or kdialog for the prompt, and a terminal (e.g. ptyxis or whatever you like) for streaming output and allowing subsequent prompts with memory. Markdown formatting (and formatting of thinking tags) with deno + glow.
- Ollama (with any model installed, e.g. gpt-oss or gemma3, etc.)
- Deno (for running glow with streaming)
- glow (for markdown formatting output)
- gum (allowing multi-line input in the prompt, or using your favorite text editor to prompt)
- wofi, or kdialog or rofi (for the dialog prompt. wofi is default, you can modify a single line (PROMPT variable) in
ask_llm.shwith your dialog tool of choice) - a terminal of your choice (for the output.
footis default, but you can modify a single line (the last one) inask_llm.shwith your terminal of choice) - stdbuf (for streaming)
- tee
Move render_stream.ts and ask_llm.sh to /usr/local/bin and make them executable with chmod +x.
Then add a keyboard shortcut to launch ask_llm.sh, for example SUPER+Backspace, in your DE of choice (e.g. KDE).
You can change the model name in ask_llm.sh (default is gpt-oss:20b) to whatever you want to use through ollama.
You can change the dialog used for the prompt, for example rofi or kdialog (wofi is default). Just change the PROMPT variable in ask_llm.sh to whatever. Same for the terminal (last line of ask_llm.sh).
If you use wofi, you can copy-paste the contents of .config/wofi to your home folder to make it look like the wofi screenshot below. In style.css, change <ABSOLUTE_PATH_TO_YOUR_HOME_DIR> to your home dir, e.g. /home/myuser.
wofi version of prompt dialog:
Terminal output (showing multi-line input in the prompt, support for Magistral thinking tags). Foot is default terminal but can of course be modified to your favorite terminal of use in the ask_llm.sh script:
Updated version with now showing the user's prompt in the output:
