Plugins

LLM auto-approval for sandbox escalation requests, with presets and a fail-closed fallback.

QQ 2007-inspired DSH Web skin: 72 native theme tokens, scoped three-pane chrome, original offline art and pixel buddy, an optional synthesized send chime, responsive/accessibility fallbacks, and reversible Settings switches.

Real-time token billing for DSH web: official DeepSeek CNY pricing with automatic peak/off-peak switching, hourly price sync from the official site, visual custom model prices, and multi-currency fallback.

The model switch for DSH: point it at any OpenAI-compatible endpoint, with curated free/cheap DeepSeek provider presets and automatic fallback when a free tier rate-limits you.

Web reading tools for text-only agents. Five read-only tools: web_fetch (any URL rendered as clean Markdown or plain text), web_links (inventory of a page's links), web_feed (RSS 2.0 / Atom entries), web_headers (HTTP status, response headers and redirect chain without downloading the body, HEAD with automatic GET fallback), and web_table (HTML tables extracted as structured rows with header detection and colspan/rowspan grid expansion). Transport is Node fetch plus a zero-dependency HTTP proxy implementation (CONNECT tunnelling, absolute-URI requests, NO_PROXY with suffix/wildcard/IPv4-CIDR matching, Proxy-Authorization). Zero runtime dependencies.

DeepSeek Harness (DSH) 推理等级自动选择插件：由 Jev System One 模型判断每条消息值多少思考量，按模型声明的等级自动推导档位，上下文信封让「继续」这类追问继承话题深度，低置信度向上取，任何失败都静默沿用原等级。 | Jev-driven reasoning effort per message: per-model ladders derived from what each model advertises, a fixed-size context envelope so follow-ups inherit topic depth, ties break upward, silent fallback on every failure path.

Live dot-matrix whale wallpaper for the DSH Web UI: ~1,750 breathing WebGL2 particles with pointer liquid vortices, light/dark themes, and a Canvas2D fallback.

Subtitle-first video transcription with SRT export, cancellable job controls, and a faster-whisper large-v3 fallback when subtitles are unavailable.

DSH-native uv environment manager: create, sync and run Python virtualenvs from the settings tab, with pip fallback, read-only conda visibility and a dry-run project sync preview.

Layered long-term memory MCP server: raw turns → task fragments → daily summaries → topic indexes, local MiniLM 384-dim embeddings or OpenAI-compatible API backend, semantic + Jaccard-fallback search, one-command DSH plugin install.

Free web search with 10 engines (Bing/DuckDuckGo/SearXNG/AnySearch free + Exa/Tavily/Keenable/Perplexity/DeepSeek paid), automatic fallback chain, time-filtered advanced search, platform search (GitHub/Reddit), web page fetching, LRU caching, and a settings UI. No API keys required for basic use.

Adds 20 locales (Traditional Chinese HK/TW, Japanese, Korean, French, German, Spanish, Portuguese, Italian, Russian, Ukrainian, Polish, Dutch, Turkish, Arabic, Hindi, Indonesian, Vietnamese, Thai, Swedish) to the DeepSeek Harness Web UI, with hand-polished translations, runtime fallbacks, and automatic translation of market and third-party text.

Aggregated web-search provider for the built-in web_search tool: one prioritized queue over nine providers (AnySearch, TinyFish, Tavily, Brave, Exa, Firecrawl, Jina, SerpApi, Serper), with multi-key rotation per provider, ordered fallback on failed attempts, and a live settings card.

AI-native music player plugin for DeepSeek Harness: aggregates NetEase Cloud Music, QQ Music and KuGou search with cross-provider playback fallback, renders word-by-word karaoke lyrics synced live over USB HID to HALO PIXELBAR speakers, and exposes search, playback, queue, alarm and speaker control as model tools.

Tavily Search API as a web search provider (settings toggle with DeepSeek fallback, keyless search) plus a Tavily MCP bridge: keyless search and extract tools, crawl and map with an API key.

Speaks agent replies in the DeepSeek Harness web UI through a provider fallback chain (OpenAI, ElevenLabs, Google, Azure, Groq, Deepgram, OpenRouter, Edge, Piper, eSpeak), so a failing or rate-limited provider falls through to the next instead of going silent.

Route OpenRouter requests through a configured provider list and quantization cap, injected as provider.only/order with allow_fallbacks, plus provider.quantizations.

Native image attachments for text-only DeepSeek in the Web GUI: pasted or dropped images appear as thumbnails in the session, and before dispatch the host reads them with the free Zhipu GLM-4V-Flash vision API (glm-4v-flash fallback chain) and substitutes the description, so DeepSeek answers about the image while the original is kept in history.

Model-facing `vision` tool for DeepSeek Harness: describe and OCR image files by calling the free Zhipu GLM vision API directly (glm-4v-flash fallback chain), no external CLI required.

DeepSeek Harness (dsh) plugin: live NVIDIA GPU monitor in the web rightbar. NVML-first on Linux (nvidia-smi fallback). NVIDIA only! Not AMD/Intel/macOS. Multi-GPU util/VRAM/power/temp, processes, sparklines. Installable dsh.bundle.

TypeSafe Jev (System One decision model) for DeepSeek Harness, end to end: jev_decision (choice/score/noul, parallel, typed answers + confidence), a Settings > Plugins card for the API key and switches, inline tool views showing every decision in the conversation (answer, confidence, latency, cost), an auto-guard (deterministic + Jev risk/loop checks), a local /jev dashboard, and an online verification benchmark (96.3% on 27 labeled questions, median ~283 ms on jev-latest, 2026-09-21). Honest by design: real API only, no mock fallback.

Compaction fallback for compaction-basic: when its LLM summary throws, comes back empty or cannot fit the window, the span is compiled deterministically with a seq pointer on every line (vendored dsh-compaction-instant compiler). Adds recall and search tools, /recall and /context commands, an idle trigger at 45% of the window and an on-disk span archive. The engine mounts as one row in the agent preset (docs/dsh-integration.md).

Live Android device pane next to your DSH conversation: scrcpy-server H.264 streaming (WebCodecs, 30-60 fps), agent tap/swipe/text with auto-screenshot verification, uiautomator element tree with tap coordinates, host-managed recording + logcat debug sessions. FLAG_SECURE pages are detected automatically with a mitigation chain and a tappable text-view fallback. Emulators and USB real phones, human+agent co-operation, session-isolated, loopback-only.

Model capability editor: edit each model reasoning-effort levels and image-input (multimodal) declarations inline in the official model page, written back to settings.yaml as a whole group, with a floating entry as fallback when the anchor breaks.