Player not loading? Watch on YouTube
OpenWhispr is presented as an open source, system-wide voice dictation app for Mac, Windows and Linux. The speaker demonstrates text entry in a browser and spreadsheet, email formatting, and brainstorming in Obsidian. In his setup, Super V starts and stops dictation, then pastes the result into the focused text field. Super Shift V invokes the assistant with selected text as context. Examples include converting text to uppercase and rewriting a Python conditional with match case.
The setup separates speech recognition from the language model used for assistant requests. The speaker uses OpenAI for both, but explains that Ollama on localhost can supply a local LLM for the assistant. Speech recognition also supports a custom endpoint with a server URL and model name, providing a self-hosted option. The demonstrated configuration uses cloud APIs; it does not establish offline operation.
Linux installation options include Debian and RPM packages, tar.gz, AppImage and an AUR package. The speaker reports that basic dictation worked out of the box on his Wayland system, while assistant mode and shortcuts required patches and Hyprland key bindings. He also says the ready-to-use free version restricts MCP and API integrations to a paid plan. He describes the project as MIT licensed and discusses modifying a fork to add integrations or fix compatibility issues.
