dsh-open-eyes

<h1 align="center">dsh-open-eyes</h1> <p align="center"><strong>Let the multimodal model you choose become DeepSeek's eyes.</strong></p> <p align="center">English · <a href="./README.zh-CN.md">中文</a></p>

What it does

The main model used by DeepSeek Harness does not always support images. When a conversation involves a screenshot, photo, chart, or interface, dsh-open-eyes can send the image to a separately configured multimodal model and return its analysis as text to the current conversation.

The analysis remains part of the same conversation as ordinary text.

If the current main model already supports images, the plugin stays out of the way and DSH keeps using its native image path. Images pasted, dropped, or selected in the WebUI are bridged only when the current model is explicitly known not to support images.

The plugin also provides the vision_analyze tool for analyzing local image paths and explicitly enabled remote image URLs.

Unofficial community plugin: dsh-open-eyes is an independent community project. It is not affiliated with, endorsed by, or maintained by DeepSeek.

Three provider API formats are supported:

• OpenAI Responses
• OpenAI Chat Completions
• Anthropic Messages

Any vision service that implements one of these APIs can be connected. You choose the endpoint, model, and credentials; the plugin is not tied to a particular provider.

API keys are resolved through DSH Credential References and should never be placed in cordis.patch.yml, conversation text, or tool arguments. Local images are workspace-contained by default, and remote image URLs are disabled by default.

Images handled by the bridge are sent to the third-party provider you configure. Review that provider's retention, privacy, and billing terms before use. Treat the vision model's response as evidence, not as instructions to execute.

Usage

Requirements:

• Node.js >=22.19.0
• DeepSeek Harness 0.1.0-rc.6

DSH profiles are independent. Install and configure the plugin separately in each profile where it is needed.

Option 1: ask another harness to install it

Any harness with Shell access and permission to manage local DSH profiles can perform the installation. Prefer a harness other than the DSH instance being modified: installing or updating the active DSH profile may require a restart and interrupt the current DSH task. If the harness cannot run local commands or edit the profile, use the manual steps below instead.

Give the other harness this prompt:

Install dsh-open-eyes in my local DeepSeek Harness web profile and configure
vision-bridge.

Use the DeepSeek Harness plugin command:
dsh plugin --profile web add dsh-open-eyes

Use only a Credential Reference in the configuration. Do not put an API key
in YAML or in the conversation. Preserve the existing profile configuration
and do not change unrelated rows.

When finished, run:
dsh --profile web --dump-config

Confirm that vision-bridge and vision-bridge-skill are loaded, then remind me
to restart dsh web and reload the page. If changing the active profile would
interrupt this task, stop after verification and let me restart it myself.

Option 2: install it yourself

Install from npm:

dsh plugin --profile web add dsh-open-eyes

You can also install the tarball attached to a GitHub Release:

dsh plugin --profile web add ./dsh-open-eyes-0.1.0.tgz

…