Include file contents in llm chat

[jritsema]

Curious, is there a way to inject the contents of a file while in the middle of a chat? If not, I'd love to be able to do something like !file foo.txt

[spm1001]

Similarly I'd love to be able to start a chat with a corpus of input (text files, images), plus an initial prompt, then continue the conversation, but cat Screenshot\ 2024-12-12\ at\ 14.06.20.png | llm chat "describe this image and then let's talk about it" -a - -m gemini-2.0-flash-exp gives. (I'm assuming/hoping this will mean that I don't have to re-transmit the image for each 'turn' in the chat.)

Usage: llm chat [OPTIONS]
Try 'llm chat --help' for help.

Error: No such option: -a

[Grynn]

Something like @ instead of !file would be cleaner I think
Copilot in VSCode & Cursor do this, meaning they let you type @ to attach a file.

GitHub's issue editor uses # to reference issues and @ to reference people.

One way to do this might be an additional hook: prepare-prompt, which allows plugins to modify a prompt before it is executed by the model. The context could/should include file attachments, model id etc. Something like:

context:
  attachments:
    - { filename, mimeType, handle }
  model-id:
  model-name:
  prompt:
  conversation id: #if continuing an existing conversation.

You'd need handle & mimetype for when the input is not a file on disk, but a pipe.

Then people can make and activate their own plugins like:

• llm-ctx-attach-files

[fakio]

working on something like this on #734

[jritsema]

@Grynn I like this idea

Something like @ instead of !file would be cleaner I think Copilot in VSCode & Cursor do this, meaning they let you type @ to attach a file.