dsh-image-amnesia
Run the following command in DeepSeek Harness:
dsh plugin install Choco-Zz/dsh-image-amnesia
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install Choco-Zz/dsh-image-amnesia in the DeepSeek Harness terminal to install; the source is available at https://github.com/Choco-Zz/dsh-image-amnesia
About this plugin
When routing vision-capable models (Grok, Claude, GPT, etc.) through OpenAI-compatible relay gateways via DSH, every image you paste or read lingers in history and gets re-uploaded on each subsequent turn. Most relay gateways hit their throughput ceiling long before DSH's default 20 MB / 600-image visual offload budget, causing connections to drop the longer the session runs. dsh-image-amnesia was built for exactly this scenario: it projects the outbound request by dropping the oldest images first while always keeping the newest N images, so the current visual context still reaches the native vision model untouched.
The plugin takes effect per profile with a single install—main agents, subagents, task-board cron jobs, and new chats all share the same wrapping logic automatically. Local session files retain every original image in full; only the copy sent to the relay is trimmed. Defaults are sensible (6 images, 6 MB byte budget, newest 1 image never dropped), and all knobs are adjustable in settings.yaml.
If you regularly paste multiple images into a single session and rely on relay gateways rather than direct API access, this plugin meaningfully cuts your disconnect risk without sacrificing visual understanding of the most recent content.
Use Cases
- Frequent relay disconnects after pasting multiple images in one session
- Throughput bottleneck from re-uploading historical images on every turn in long chats
- Unbounded image accumulation across subagents and task-board jobs sharing a profile
Best For
- Users who regularly paste multiple screenshots or photos in a single session
- Developers calling vision models through OpenAI-compatible relay gateways
- Teams managing multi-agent workflows at the profile level
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.