Plugins

Remembers the model used before quitting and restores it as the startup default.

Connect DeepSeek Harness to the local Codex CLI with dynamic model discovery, image input, Harness-owned tools, and read-only ChatGPT plan and quota display.

After user setup, guides native subagents toward model preferences by task type, using native model-selection permissions with no extra routing-model calls.

Local LoRA and QLoRA fine-tuning in a panel: import JSONL, see a VRAM estimate before training starts, evaluate, and export an adapter. One-click Python venv setup, but you must supply your own Python 3.10+ and an NVIDIA GPU (6 GB or more). Verified end to end on Windows 11 with an RTX 3060 6 GB; the installer refuses non-Windows hosts, so Linux, macOS and other GPUs are unsupported for now.

Adds MiniMax open-platform models to the DSH model picker: paste the API key in Settings and MiniMax models become selectable, with the M Plan quota windows shown beside the composer.

Signs in to ima.qq.com (Tencent IMA) with a WeChat QR code and runs a localhost OpenAI-compatible relay in front of its SSE chat endpoint, so the four models behind it appear in the picker.

TeleAgent (China Telecom TeleAI "Stellar" agent) model provider for DeepSeek Harness with the working v1 gateway signing: two-round HMAC-SHA256 keyed by the login JWT signature segment, which fixes the 40104 rejections the older local-v1 signing produced. Signs in through a dedicated browser window over the DevTools protocol — the plugin intercepts the login callback the same way the official desktop app does, so nothing has to be copied by hand — and also accepts a pasted token. Ships points balance, daily check-in, an in-composer account chip, and a configurable context window (100K by default, 16K–2M, since the gateway declares no catalogue and the official client ships one shared 100K preset for all three tiers). Cannot be installed alongside anye-99/dsh-codearts-auth: both register the same teleagent provider, and this plugin reuses its credentials and derives its SSE/tool-call parsing from it.

QUASAR — a space-themed skin for DeepSeek Harness: an ink black hole in the sidebar and a live accretion disk rendered behind the app, with frosted-glass panels. Motion is canvas-driven and honors prefers-reduced-motion.

Global cyber-terminal theme: a click-through full-screen Matrix code-rain overlay, phosphor-green UI styling, a sidebar one-click toggle, and adjustable speed, density, opacity and glow with three palettes.

A free, unofficial Cyrene-inspired theme for DeepSeek Harness with three palettes, a splash screen, and a chat watermark. Code is MIT; character fan art is excluded from MIT.

Drag the two handles to resize and move the DSH chat area as a floating window; position and size persist across reloads, and movement is clamped to the conversation pane so it never leaves the screen.

Wallpaper layer behind the whole DeepSeek Harness Web UI, fed by image URLs, local files and folders, or a Wallpaper Engine workshop directory, with opacity, blur and translucent-panel sliders, a slideshow with random order and cross-fade, muted video playback, previous/next keyboard shortcuts, and a per-URL archive that keeps every fetched picture under "Loaded images".

Per-turn cost in CNY, with 本轮 / 今日 / 会话 totals shown above the composer. Tokens are priced by model and by the moment they were spent (peak hours and holidays); native DeepSeek search spend can be folded in when the optional audit plugin is installed.

Read-only observer of native DeepSeek search model calls: appends each call token usage to a JSONL file so accounting plugins can price it. Requests and responses are not modified.

Balance strip in the composer dock — a green-to-red OKLCH ramp driven by your DeepSeek balance, with a dynamic toggle, texture options and a built-in settings panel.

Makes the agent narrate progress during long tool chains instead of staying silent until the final answer: a standing system-prompt rule that travels with every request, plus a notice injected after every N tool calls. A panel on the plugin page tunes the cadence, the reminder budget, and both texts; changes apply live.

Shows the cost of the current session's latest exchange, meaning everything since your last message, as a CNY amount to the right of the permission control in the composer, calculated from provider-reported token usage at DeepSeek's peak and off-peak prices.

Speaks the estimated spend of each turn aloud with an Alipay-style voice clip or a bundled meme voice; amounts come from dsh-cost-meter when installed, otherwise from per-model token prices. All audio is bundled, so nothing is fetched at runtime.

Cross-session per-model token usage replayed from the harness's own session logs, with a usage panel tab (requires dsh-better-sidebar) and a per-conversation model popover.

DeepSeek balance and today's spend from the official usage export, in one line at the right end of the DSH sidebar account row; hovering opens a card with requests, tokens and the peak/off-peak countdown.

Queues composer messages sent during DeepSeek peak billing hours and delivers them 3 s after the next off-peak window starts. A dock above the composer shows the countdown with send-now / edit / cancel, the queue survives closing the browser or restarting DSH, and attachments are staged on the host.

One Settings page for OpenCode Go: an account pool that rotates A-to-B when a key runs out or stops authenticating, per-account 5-hour/weekly/monthly allowance, per-model spend with remaining quota, capability tags (image, video, text-only, tools, reasoning), a live OpenCode-vs-official-API cost comparison, and 10 UI languages.

A countdown pill under the message composer showing how long is left before you clock off, with a workday progress bar; click it to set your own start and end times (saved in this browser), and after the end time it turns into an overtime timer.

Right-click any selected assistant reply and add it to the composer as a removable quote chip. The selected text is delivered to the model as hidden context, so your message bubble keeps only your own words.