Introduction

DeepSeek Harness (DSH) enables developers to build agents, but natively supports only text-based interaction. The imgpost plugin removes this limitation by allowing DSH to send images during conversations and to call vision APIs to understand image content.

Overview

imgpost is a client plugin maintained by hige6 and licensed under MIT. Through several tool functions, it enables DSH to process local images, web page links, AI-generated images, and perform external visual recognition.

Core Features

  • send_image: Sends local files / web URLs / base64 images into the conversation as persistent attached media.
  • generate_image: Calls any OpenAI-compatible /images/generations endpoint to generate an image and sends it directly.
  • imgpost_read_image: Uses an external vision API to read image content — such as OCR text, scene details, and layout; results are cached to disk by image fingerprint.
  • imgpost_check_backend: Reports whether the baseURL / model / key for the image-generation endpoint are all present, and clearly states which item is missing.
  • vision provider wrap: A transparent pass-through that automatically wraps a vision channel around text-only models, allowing any model to “see” images.
  • Settings management: Manages image-recognition backends (primary/backup) and the upstream wrap list directly in DSH settings; changes take effect immediately after saving.

Installation

Use the official script to install it.

powershell -ExecutionPolicy Bypass -File .\scripts\install.ps1

Configuration and Usage

  • Configuration file: Located at ~/.dsh/vision-sender.
  • Default configuration: The official DeepSeek model is preselected by default.
  • Security note: The apiKey is never sent back to the browser; the API includes built-in access controls (custom headers and same-origin checks).
  • Dependencies: Depends on @deepseek-ai/cordis: ^4.0.1, with zero third-party dependencies.

Notes

  • The plugin runs with the current DSH process permissions; review the source code and license before installing.
  • The routes registered by the plugin operate outside the host’s token authentication, but the API endpoints implement three layers of validation.

Closing

For more details, see the GitHub repository: https://github.com/hige6/imgpost