dsh-file-upload

DeepSeek Harness natively drops dragged images whenever the active model lacks vision capabilities. This plugin eliminates that friction by introducing a unified upload button and seamless drag-and-drop zone directly within the composer toolbar. Its intelligent routing engine handles the heavy lifting: when your current model accepts images, files are routed straight to the official attachment rail, leveraging the DeepSeek Files API for automatic ID reuse and optimized delivery. If vision support is disabled or not available, it gracefully degrades to embedding a clean file path in your draft, bypassing native image blocks entirely while ensuring any baseline model can still parse the reference.

Beyond images, the tool fully embraces complex file management. Dropping an entire folder recursively reconstructs its structure under a configurable local archive, leaving only the original path in the draft for clean context handling. A dedicated settings panel puts you in control, allowing toggles for the official attachment channel, custom path prefixes, clipboard interception modes, and strict same-origin security defaults—all without requiring external system dependencies thanks to its pure Node.js implementation.

Designed for developers, researchers, and power users who frequently inject external resources into AI conversations. Whether you’re optimizing multimodal workflows by maximizing the official attachment rail, managing large codebases or datasets via structured folder archiving, or simply needing a reliable fallback mechanism when switching between chat models, this plugin streamlines how files bridge the gap between your local machine and the harness interface.