Skip to content

Run transcription / selected text through LLM Agent #136

@stromseng

Description

@stromseng

Describe the solution you'd like
If you hold down a specific hotkey/key combination when transcribing, one of two things shall happen:

  • No selected text: The transcription shall be passed to an LLM agent to handle, and paste resulting text.
  • Selected text: The transcription shall be instructions for how an LLM agent should transform or do stuff with the selected text, then paste the resulting text.

I propose the following implementation:
Allow the user to set an arbitrary terminal command/script that shall handle the transcribed words and output the transformed text.
This way, the user can choose which LLM agent they want to use themselves, as long as it has a CLI.
For example, using Claude Code you could run its CLI with the structured argument flags and parse it using jq to get your transformed text output.

Describe alternatives you've considered
See #62 for a similar proposal. Although i believe my suggestion is more powerful, although a bit different. This could be a nice complement to it.

This could enhance #121 . We would allow for shell scripts in the transformation pipeline, and make sure to provide the selectedText to the pipeline as well.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions