Back to home@johndfowler

dsh-photos

No description

Stars
0
Language
JavaScript
Created
Aug 29, 2026
Updated
Aug 30, 2026
GitHub repo

Introduction

dsh-photos

Photo upload for the DeepSeek Harness composer.

A 📷 button floats next to the voice mic at the composer's top-right. Tap it and you get the native picker — on iPhone: Photo Library / Take Photo / Choose File; on desktop: a file dialog. Picked images land in the composer as draft attachments with thumbnails, ready to send.

How it works

The stock composer already supports image attachments via paste and drag-and-drop — there was just no way to initiate that from a phone. This plugin feeds the picked files into the composer's own paste pipeline (a synthetic ClipboardEvent with a DataTransfer), so:

  • stock validation applies (media types, per-message count, byte limits, with the stock toast errors)
  • stock thumbnails render in the attachments row
  • the stock send flow serializes them — identical to a real clipboard paste

No React internals touched, no DOM mutation inside React's tree (the button and file input are fixed/hidden overlays).

Install

cd ~/.dsh/profiles/web
pnpm add file:~/path/to/dsh-photos

Add dsh-photos to the profile's dsh.profile.bundles in package.json, restart the GUI, and the 📷 appears.

Notes

  • Sending images needs a vision-capable model selected (a GLM-5V variant, Gemini, Qwen-vl, Solar Pro 4, …) — admission is model-agnostic, but the model must accept image input for the turn to succeed
  • Multiple selection works (the picker allows it; files arrive as one paste)

License

MIT