dsh-plugin-vision-fallback
Run the following command in DeepSeek Harness:
dsh plugin install WEIHAOLEE/dsh-plugin-vision-fallback
Paste the following prompt into your AI chat to install this plugin:
Run the install command in the DeepSeek Harness shell to add this plugin; the full source repository is at https://github.com/WEIHAOLEE/dsh-plugin-vision-fallback , and the Visual Agent toggle will appear next to the input box after installation.
About this plugin
Most of your DSH sessions run on text-only models like Flash or Pro, but every now and then you need the agent to actually look at a screenshot or a chart. dsh-plugin-vision-fallback adds a small toggle next the input box: when it is on, image-bearing messages are routed to your configured vision model while plain-text requests keep flowing to whichever model you originally picked. Flip the toggle off and everything snaps back instantly.
Under the hood the plugin temporarily hijacks the session route onto a registered vision-fallback channel, remembering your original model and restoring it on toggle-off. Each vision call sends only the latest message (one image plus its caption) with no conversation history and no compression, so you stay within token budgets. It ships with Volcano Ark Doubao as the default provider but works with any OpenAI-compatible endpoint. The Settings panel lets you add, remove, and test multiple providers with hot-reload, so no restart is needed.
This plugin fits DSH users whose workflow is primarily text but who occasionally need an image check, especially those juggling several vision providers at once. Keep in mind it is still in an early, experimentally-verified state, validated on a single Windows machine with one provider pairing. Edge cases such as concurrent multi-image reads, sub-agent sessions, and very large uploads have not been fully stress-tested. Evaluate it on your own stack before relying on it in production.
Use Cases
- Occasionally needing the agent to inspect a screenshot or chart during a text-based session
- Managing multiple vision providers with hot-switching and no restart required
- Avoiding a separate vision-model session for the occasional image check while keeping the current text model
Best For
- DSH users whose workflow is primarily text but who occasionally need image inspection
- Workflow developers connecting multiple OpenAI-compatible vision providers
- Users who want to keep their existing text-model setup untouched while adding on-demand vision capability
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.