There are a lot of tools already that do local speech-to-text, but none of them allowed me to "select" code/text in a file and then paste a reference to my selected code together with the transcript.
I wanted this to work locally and I wanted it to work independently of the IDE (because I'm using multiple). I also wanted it to work with any coding agent (Claude, Codex, OpenCode).
So I put together parakeet & ripgrep in a Rust tauri app. It works perfectly for me and I'm using it every day whenever I'm coding. Maybe it's also useful for others.
Nice stuff. I keep wanting to be more speech oriented but it's been hard to get myself to move to that. I like how you use the accessibility APIs to get data out; that feels like such crucial leverage point for expanding agency!
> When you press the hotkey, macOS Accessibility APIs read the selected text and line numbers from the active editor window. If the Accessibility API doesn't return selected text (some apps don't support it), Voxcode falls back to simulating Cmd+C.
Regarding working with code, I've quite enjoyed using the mcp-neovim-server. Asking it to show & select pertinent lines can be helpful, especially when there's multiple file references at once. Turning this around & asking the LLM to get data out would be possible, but as you say, this would be IDE specific. Ideally there would be some good cross-IDE MCP tools.
https://github.com/bigcodegen/mcp-neovim-server
There are a lot of tools already that do local speech-to-text, but none of them allowed me to "select" code/text in a file and then paste a reference to my selected code together with the transcript.
I wanted this to work locally and I wanted it to work independently of the IDE (because I'm using multiple). I also wanted it to work with any coding agent (Claude, Codex, OpenCode).
So I put together parakeet & ripgrep in a Rust tauri app. It works perfectly for me and I'm using it every day whenever I'm coding. Maybe it's also useful for others.
Nice stuff. I keep wanting to be more speech oriented but it's been hard to get myself to move to that. I like how you use the accessibility APIs to get data out; that feels like such crucial leverage point for expanding agency!
> When you press the hotkey, macOS Accessibility APIs read the selected text and line numbers from the active editor window. If the Accessibility API doesn't return selected text (some apps don't support it), Voxcode falls back to simulating Cmd+C.
Regarding working with code, I've quite enjoyed using the mcp-neovim-server. Asking it to show & select pertinent lines can be helpful, especially when there's multiple file references at once. Turning this around & asking the LLM to get data out would be possible, but as you say, this would be IDE specific. Ideally there would be some good cross-IDE MCP tools. https://github.com/bigcodegen/mcp-neovim-server