플러그인

Node-based GLSL workbench: wire a shader graph on a canvas and preview it live, compile it to Shadertoy-compatible GLSL with multi-pass buffers (BufferA–D), hot-reload as you edit, cross-compile to HLSL / MSL / SPIR-V, and keep projects, materials, snapshots and resources (images, audio, lyrics) in a local workspace. Ships 57 built-in nodes, 8 bundled examples and 11 agent tools. No runtime dependencies.

Re-encodes WebP attachment images to PNG so alpha-channel images load on LLM backends that cannot decode WebP (e.g. llama.cpp).

Talk to DeepSeek Harness from Feishu / Lark. A chat maps to a DSH session, messages become turns and answers are pushed back as cards; questions, permission requests, model switches and conversation switches are interactive cards, and inbound files and images are downloaded into the workspace.

Local OCR with nothing sent to a cloud: mounts the PP-OCRv6 Medium stdio MCP server (PaddleOCR on ONNX Runtime, CPU) as mcp__ppocr__ocr_image and mcp__ppocr__ocr_pdf for images and multi-page PDFs.

Shows oversized images to the model at readable resolution instead of one downscaled copy: before admission, a large upload is split into as many tiles as the attachment store allows (an 8000x8000 screenshot becomes 20 tiles of 2000x1600), so small text survives. Replaces the local attachment store; adds no settings of its own.

Displays generated images in a conversation gallery with a lightbox viewer.

DeepSeek Harness plugin: a model-facing `chrome` tool that drives the installed Google Chrome over CDP — screenshots returned as images, DOM/text/eval, console & network diagnostics, real input events. Runs in the Host, so no per-call approval and no bash sandbox limits.

Inline image chips for the DeepSeek Harness composer — paste an image and it lands in the text as a numbered chip (image-1.png) with hover preview. Images still ship through the official attachment pipeline, names restart per message, the dock and sent-message thumbnails carry a sequence badge, and a pre-send self-check warns when the names in the text no longer match the images in that message.

Generate and edit images, generate video, and cut video locally through a GenBox server, with a zero-key test bench.

Kite-flying engine: while the agent codes, an animated kite drifts and sways in the wind on screen, tethered to the bottom edge of the window — the busier the agent, the denser the events and the higher it flies; ships a Weifang-style framework card deck (sand-swallow, goldfish, butterfly, bagua, dragon-head and more, hard-wing / soft-wing / flat / dimensional frames), with shape x pattern x colors all swappable data configs, and supports pasting user images onto the kite face with real-time affine transforms as the kite banks.

AI-native expressions for two-way chat: people see images, while text-only models read fixed meanings; includes a picker, custom expressions, import/export, and two visual styles.

DeepSeek Harness plugin for Xiaohe Canvas: AI-powered e-commerce content creation (images, videos, workflows).

PDF electronic signing: generate a handwritten-style signature image, stamp images and typed blocks onto PDF pages, and apply a PKCS#7/CMS digital signature with a PKCS#12 certificate you supply. Stamping first keys an opaque background out to transparency - un-compositing each pixel so the ink keeps its original colour - and trims to the ink, so a seal stamped over a signature does not erase it.

WeChat bridge for DeepSeek Harness over Tencent's iLink bot channel: one session per contact, inbound text/images/files, and WeChat's native typing indicator while a turn runs.

Computer Use for macOS desktop applications: reads any app's accessibility tree as text with element indices, then clicks, types, pastes, scrolls, drags, selects text and sets values through it, plus screenshots read back as images. Driven by a persistent JavaScript REPL — the agent gets two tools (`js`, `js_reset`) and calls an `sky` API inside them, so state survives between calls and the tool schema stays fixed as capabilities grow. Accessibility trees come back as diffs after the first read; actions address elements by index and indices from a stale read are refused rather than misapplied; typing waits for keyboard focus and reads the field back; paste restores the clipboard. Requires macOS 14.4+, Accessibility permission (Screen Recording only for screenshots) and Node 20+; ships a prebuilt native module, so nothing compiles at install time.

Sidebar button that opens a semi-transparent popup for browsing, previewing, downloading and deleting generated images and videos, with a settings page for the media, record and download directories.

Sidebar notebook with titled notes and inline images (videos rejected); click a title to copy its body to the clipboard, reference a note in the composer with @, and optionally open the notebook for every new session.

DeepSeek Harness Web plugin that keeps image-bearing Tool results visible after Compact transcript folding

Digests the images in a prompt with a vision model before admission, so any model — a text-only one included — can read them without the session ever switching models, and adds a describe_image tool for image paths.

Trims historical images in outgoing chat requests down to a recent-image count, learns the provider per-prompt image cap from HTTP 400 responses, and retries with fewer images so image-heavy sessions keep working.

Three tools for DeepSeek Harness: typeset LaTeX formulas into images, draw mathematical figures from a declarative spec, and convert a formula or SVG into an image ready to embed in a document.

Configure multiple image providers in Settings and call image_generate with the one selected model; images save under generate/image and show inline in the conversation.

Web management panel for Docker on DeepSeek Harness: manages Docker MCP servers (deploy/remove, official catalog search/pull, gateway start-stop-restart, custom deploy) and containers/images (list, start/stop/restart, remove, logs, inspect, interactive terminal, create, pull, build, load) across local CLI and SSH remote targets with one-click switching. Credentials are encrypted with Windows DPAPI, HTTP is loopback-only, and command parameters are whitelist-filtered. Build-free pure ESM; client is zero-dependency.

Auto-switch DeepSeek models: routes every request to the vision model when images are present, to the pro model for complex tasks, and to the fast model otherwise; includes a switch_model tool for manual overrides.