DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Automatically transcribes pasted images to text for pure-text models via a transparent twin provider, using a vision model you configure — no model switching or relay required.
LINE:HASH anchored file read and atomic batch edit for reliable multi-round model editing under line drift and concurrency.
Enable DeepSeek V4 Flash on the Volcano Ark plan API in DeepSeek Harness by overriding baseURL on the built-in provider for 10x cheaper off-peak tokens.
A community plugin that steers DeepSeek Harness reasoning into Simplified Chinese via a zero-cost guidance layer with an optional translation fallback.
Adds a project-level folder hierarchy to the DSH sidebar for grouping, renaming, searching, and archiving conversations within a project.
Adds a scenic mountain-and-sea background to the DeepSeek Harness Web UI, with daily rotation, Ken Burns slideshow, frosted-glass panels, and dark/light theme support.
Adds FriendliAI as an LLM provider to DeepSeek Harness via its OpenAI-compatible chat API, with a live-fetched model catalog at runtime.
An installable bundle that bridges the DeepSeek Harness CLI to Multica via a versioned JSONL process protocol, supporting model discovery, session execution, and MCP server mounting.
Side-by-side evaluation of multiple coding models in DeepSeek Harness: concurrent runs in isolated Git worktrees with tool trajectories, diff stats, and one-click adoption.
A 0-100 quality scorecard for dsh plugins with grade, six-component breakdown, and fix suggestions, built on npm metadata with zero runtime dependencies.
Add up to three shorthand buttons above the DSH composer to switch provider, model, and reasoning-effort combos in one click without altering agent configuration.
Attach a read-only reference library to each session so the agent consults your local directories first before falling back to external sources.