플러그인

Bundles the PrismSearch MCP server and its agent skill. Web search over Bing and Baidu with no account or API key, plus Google and Brave when API keys are set and DuckDuckGo, Sogou, 360 and Yandex when enabled in config; automatic engine fallback and an aggregate mode. Also image search, WeChat Official Account article search, and fetch_page, which returns web pages and PDFs as Markdown in segments.

An Auto toggle next to the model selector that schedules reasoning_effort per request: complexity is scored by a MiniMax/OpenAI-compatible model with a zero-dependency heuristic fallback (fallbacks not cached), effort ladders adapt to each model real effort list (low/high/max, low/high/highx/max/ultra, ...), heavy tool calls upgrade the tier within the same session, the lit button and reply badges show the effort actually used, and every decision plus observed token usage lands in a JSONL ledger. A standalone-button fork of imkingjh999/dsh-adaptive-effort; do not install it and the original in the same profile.

A local ComfyUI workbench for DeepSeek Harness (DSH): image/audio/video generation, W2/W3 sprite cards with Qwen back-view walk-up, local Flux.2 Item Studio assets with guided SDXL fallback, Hugging Face downloads, inspection and arbitrary workflows.

Sends one system notification when a DSH turn ends: per-outcome switches for completed, error and interrupted, a minimum-duration threshold, silent subagents and unfocused-window-only delivery. Works on Windows, macOS and Linux, delivered through a connected page or a native fallback (PowerShell toast, osascript, notify-send; the Linux native fallback needs libnotify). Clicking the notification brings DSH back to the front on the page/Electron channel; PowerShell and notify-send notifications are not clickable.

DeepSeek Harness tools and a Skill for md-report/1: validate and render Markdown plus declared SQLite/CSV datasets into one offline interactive HTML report or dashboard, with AntV G2 charts (ECharts fallback) and chart/table data and source-SQL views.

One-step local model onboarding plus a light harness for DSH. On load it probes loopback and private-network inference endpoints (llama.cpp, Ollama, vLLM, LM Studio), reads each model's context window and output budget from the backend, and registers the provider through settings, so a local model becomes selectable without hand-writing config. It then trims only that provider: masks the tool list through tools.restrict, drops named system-prompt sections and memory contexts, swaps the injected instruction block for a short one while preserving its first 80 and last 96 characters, and truncates tool-schema descriptions without touching structure or required fields. Cloud providers are returned as the same assembly object, untouched. The injected block is re-read from disk on every assembly, so editing the file takes effect on the next request with no reload. The panel in the plugin detail page is a live operations view: it shows whether the takeover is active and by which hook, how many characters the last assembly was trimmed from and to, and which providers are online (probed concurrently, with a real one-shot completion reporting latency and reply text per provider). If a local backend is not running, the panel offers a one-click launch: it probes the port first (never starting a duplicate), then starts the matching server (ollama serve, or a custom command you supply) detached and polls until the endpoint answers, clamping the wait to 3-600 seconds, reaping the child it just spawned if it never answers, and naming the paths it searched when it cannot find the binary. A stop button kills backends this plugin started (SIGTERM, then SIGKILL) and refuses to touch a server somebody else launched. When a model has just been downloaded and no backend exists yet, the panel can also scan the machine for .gguf files with hard caps (300 models, 20000 directory entries, Downloads only one level deep), start llama-server on the chosen one (skipping the .mmproj. vision projection and reusing a server that already answers), register the provider under the model id the backend reports, and run one real completion to prove it works. Before spawning anything it does a fit check that counts only the part of the file the GPU cannot hold: hostBytes = fileSize - VRAM, compared against max(total RAM - 2GB, free RAM). A file far larger than RAM is refused outright; a heavy host-side remainder is refused unless you confirm; the same remainder under mmap is only a warning, because the weights land in reclaimable page cache. A load-mode selector (auto / mmap / none / mlock / mmap+mlock) maps to llama.cpp --load-mode, and the default is whatever the engine picks. From 0.4.2 everything above is derived from the platform instead of assumed: home and data directories come from os.homedir() and the OS conventions (LOCALAPPDATA on Windows, ~/Library/Application Support on macOS, XDG_DATA_HOME elsewhere); backend binaries are looked up in the official install locations of each OS and can be pinned with OLLAMA_BIN, LLAMA_SERVER_BIN, VLLM_BIN or LMSTUDIO_BIN; model discovery walks DSH_LOCAL_MODELS_DIR, ~/models, ~/llama.cpp/models, ~/.cache/llama.cpp and the LM Studio data directory, with the WorkBuddy layout demoted to one candidate among them. VRAM is asked from nvidia-smi, then rocm-smi, then sysctl hw.memsize on Apple Silicon (flagged unified, so the same bytes are not charged twice), and any flip to a unified-memory budget is handled explicitly. Before building a command line the plugin runs llama-server --help once and keeps only the flags that build actually has, so older engines get --n-gpu-layers 99 and no KV quantization while newer ones get auto plus q8_0; anything it cannot probe is passed through rather than dropped. LM Studio is started with lms server start --port N rather than being lumped in with the engines that need a model argument, and vLLM gets its own --model example instead of llama.cpp syntax. Windows, macOS and Linux are all exercised in CI, including a windows-latest and a macos-latest job, because a cross-platform claim that has only ever run on one OS is not a claim. Two of those suites exist specifically to catch this machine's blind spots: one replays the whole suite with process.platform rewritten to each OS, and one deletes every advantage this development box has — no nvidia-smi, no rocm-smi, no sysctl, an empty PATH, a home directory with no models — and requires graceful degradation rather than a crash. That last one caught a real class of bug: several assertions hardcoded 'lms.exe', which cannot pass where the executable has no suffix. Measured on the reference machine: a 16.9GB 27.2B MoE with 14.5GB of VRAM holds a 285-691MB working set and sustains 76 tok/s after a 1000-2000 token warm-up, so file size alone is not a fit signal. From 0.4.3 the scanned models are shown as a table rather than a dropdown (one row per gguf, with its size and any fit warning), so you can see what is on disk before choosing. From 0.4.4 the delete button on the provider table removes the provider entry, not the weights file on disk. That distinction matters: what people actually want to clear out is the dead entry - one pointing at a port that never answered, one registered twice for the same port, one whose model file was deleted long ago - not a 15.8GB gguf. Deleting a config is undone by importing again; deleting a gguf is not. The removal goes through settings.mutate with an unset op on \['providers', id], because update merges layers: writing null there leaves a null provider key that the loader reads as a broken config rather than an absent entry, and only mutate/unset is implemented as Reflect.deleteProperty. The result is read back afterwards, because unset on a key that comes from a lower layer (the profile cordis.patch.yml or a home patch) does not delete it - it restores the inherited value, so the call succeeds while the entry stays. When the entry is still there the panel says so and points at the file to edit instead of reporting success, and a DSH whose settings service predates mutate() is told plainly rather than silently ignored. Deleting takes two clicks and clearing the confirmation is as easy as clicking another row; on success the online and baseURL caches are cleared and the tool surface is recomputed, since the provider list changed. The gguf itself is never touched by this plugin - the scanned-model table is display and selection only.VRAM reporting is driver-dependent. From 0.4.5 model discovery also reads Ollama's own storage: its blobs are named sha256-… with no extension, so extension matching could never list them. Weights are now accepted by extension OR by GGUF magic header (only dot-free files of at least 8MB, capped at 600 header reads per scan; obviously-not-weights extensions like .json/.md/.safetensors/.part are never opened). Ollama models are resolved from their manifests: the model layer is picked by mediaType application/vnd.ollama.image.model (never adapter/mmproj layers), the human-readable name is restored from the manifest path (…/qwen3/8b becomes qwen3:8b, …/latest stays bare), tags sharing one blob merge into a single row, and the metadata kv supplies name, quantization and context length. Adopting an Ollama model does not start a duplicate llama-server over the same weights: it confirms ollama serve answers on 11434 (launching it if not) and registers the provider under the tag the endpoint itself reports — the blob's sha256 name would never match anything, since 11434 lists qwen3:14b, not sha256-a8cc…. Two real-machine-only traps fixed on the way: existsSync('<file>/') returns true on Windows, so manifest-vs-directory branching now goes by parseable content instead of trailing-slash semantics; and the quantization label table had been copied from a stale ggml enum generation, labelling ggml_type 15 as Q5_K_S while ollama show reports Q4_K_M. Same-blob dedup is by (dev,ino), keeping whichever entry has a name. Verified on the reference machine: the scan goes from 1 row to 4 — qwen3:14b (8.64G) and qwen3:8b (4.87G) now appear with Q4_K_M badges — in 4ms; npm run verify exits 0 (launcher 57, portable 35, panel-host green including the linux/darwin simulated-platform runs); the new panel-host suite builds a real miniature Ollama store in tmpdir and points OLLAMA_MODELS at it, so results no longer depend on whether the test machine happens to have Ollama installed. From 0.4.6 the panel shows each model its real name instead of its file name, because some download tooling stores every weight under one generic file name, leaving the real name only inside the GGUF header. The scan now reads each file GGUF KV section - not a fixed-size head buffer, because tokenizer.ggml.tokens (hundreds of thousands of strings on a large MoE model) and the chat template push general.file_type past any window; and GGUF string arrays are interleaved (u64 length plus payload per element), so treating them as a length table walks the cursor off the rails and produces astronomically large reads. Reading and skipping are accounted separately: key names and types are always read, string values over 256KB and string arrays are only walked past without copying, non-string arrays over 64 elements are skipped outright, and a cursor past the file size stops the read, because an out-of-range position means the cursor is wrong rather than that a limit should be raised. The displayed name goes Ollama tag, then general.basename plus size_label, then general.name, then the file name, with the full string kept for the hover text since it usually carries internal build numbers; mmproj files are now identified by general.type as well as by name. A short read (truncated file) no longer feeds half a value into the parser. On the reference machine, reading one file header parses 44 keys in about 16ms, and a whole scan takes about 31ms. From 0.4.7 the smoke test no longer fails the first time and passes the second. The old code gave both request attempts one shared AbortController and one shared 30-second timer, so the total budget covered both rounds: a cold load of a large model burned it in round one, the abort left the signal permanently aborted, and the second round (the Ollama-style reasoning_effort: none retry) could not even be sent - which is exactly why the first click returned This operation was aborted and the second succeeded. Now each attempt gets its own controller and timer, a timeout retries once automatically (retryOnTimeout, on by default) because a first-request timeout almost always means the weights are still being read from disk, and non-timeout errors (refused connection, HTTP error) are not retried at all. Timeouts report themselves: the message says the wait exceeded smokeTimeout, whether a retry already happened, and points at the new smokeTimeout setting (seconds, default 60, clamped to 5-600) - port responding is not the same as tokens coming out, so it is a separate knob from launchTimeout. Unrelated: an example naming one specific model was dropped from the panel copy, since not every machine has that model downloaded, and the explanation now describes the mechanism instead. From 0.4.8 adopting a plain .gguf no longer registers an unrelated Ollama model. The port field used to follow the last click rather than the file: selecting an Ollama entry moved it to 11434, selecting a plain .gguf did not move it back, so a 11434 left over from an earlier click stayed. The server accepted it as given, probed the port, found Ollama answering and borrowed it (the skipped-launch log line about the port already answering is this path), and then the model id - derived from the file name - could not be matched against Ollama's /v1/models, where the fallback was to grab the first entry in the list. Embedding models tend to sort first, so the adopt registered a model the user never selected. Now the port follows the file: a plain .gguf landing on another engine's port (ollama 11434, vLLM 8000, LM Studio 1234) is moved to 8090 and the panel says so. Only ports belonging to a known engine are overridden - a hand-typed 9999 aimed at your own llama-server is left alone. The model-id fallback no longer guesses either: it takes the only entry when there is exactly one (a single-model backend may legitimately report a different id than the file name) and otherwise reports an error listing the ids it did see, rather than registering a wrong provider.

Routes DeepSeek models in dsh through an OpenCode Go subscription and falls back to the DeepSeek account per model when that model's allowance runs out; a plan-wide fallback covers bad credentials.

Fairy III (Zenless Zone Zero) H.D.D skin for DeepSeek Harness: boot video, an eye-core dock that can be dragged vertically so it never covers text, and time-of-day greeting voices written by a model every day (5 lines, each <= 20 characters) and synthesized locally with GPT-SoVITS, with a fallback chain of today -> latest -> bundled audio so there is always something to hear. Replies can be spoken: a trailing marker line takes priority over length heuristics, and long replies can be compressed into one sentence first. The reader hashes the utterance it just played (MD5) so each reply is voiced exactly once, and switching sessions stays silent. Personal fork of dsh-fairy-zzz with its own package, route and storage namespace, plus a rate-limited summarize endpoint, POST-only engine control and bounded TTS and greeting caches. Two diagnostic logs (client revision, decision trace) are written next to the plugin and stay on the local machine. A personal fork of AssassinVIX/fairy-zzz-plugin (upstream package dsh-fairy-zzz): install this fork or the upstream, not both - both mount the full-screen skin on the same DSH surfaces. Dynamic greetings and spoken summaries need a DeepSeek-compatible API key in ~/.dsh/fairy-tts.json (or DEEPSEEK_API_KEY); without it only the bundled greeting audio plays.

주제·키워드·DOI로 Europe PMC, PubMed, OpenAlex, Crossref에서 의학 및 생명과학 문헌을 검색합니다. Unpaywall/OpenAlex/Europe PMC로 오픈 액세스를 확인하고 OA 경로가 있으면 매직 바이트와 페이지 수를 검증해 PDF를 다운로드합니다. 없으면 정확한 실패 이유와 네 가지 수동 대안(AbleSci assistance, Sci-Hub, SciDown, X-MOL)을 기록한 결정 카드를 작성합니다. 포함된 Python 도구는 표준 라이브러리만 사용합니다.

macOS용 완전 로컬 푸시 투 토크 음성 입력입니다. 전역 단축키를 누르고 말한 뒤 놓으면 받아쓰기가 커서 위치에 붙여넣어집니다. 로컬 우선 STT(Qwen3-ASR 0.6B 4bit on MLX/Metal 기본, SenseVoice와 mlx-whisper 대체), 외부 API 미사용, launchd 자체 복구 서비스, 8항목 진단 스크립트를 제공합니다. 내장 Skill은 에이전트가 전체 도구 체인을 설치·설정(Karabiner 단축키 연결, launchd, 엔진 전환)하고 문제를 해결하도록 합니다.

컨텍스트 한도에 도달해 내장 압축이 포기되면 자동으로 한 번 더 압축한 뒤 다시 시도합니다. 그래도 실패하면 원래 세션을 핸드오프 패키지로 정리해 새 세션에서 이어갑니다. 엔진 교체와 UI 추가는 없습니다. 세션 재개는 기본적으로 명시적으로 실행하며 원래 세션은 읽기 전용입니다.

모델 오류가 나면 백업 모델로 자동 전환해 같은 단계를 다시 시도합니다. 설정 페이지의 모델 목록에서 백업 모델을 선택할 수 있습니다.

DSML tool-call 마크업이 reasoning에 새어 나와 멈춘 턴을 자동으로 계속합니다.

모델이 만든 DATE | CATEGORY | NAME 형식으로 모든 새 DSH 세션의 이름을 지정하며 결정론적 로컬 대체 방식도 제공합니다.

입력창의 모델 버튼을 제공자별 모델 목록, 검색, 고정 즐겨찾기, 모델이 제공하는 단계마다 눈금 하나인 reasoning-effort 슬라이더 패널로 교체합니다. 단계가 높을수록 aurora 그라데이션과 빛남이 강해집니다. 모델별 마지막 effort를 복원하고 기본 선택기도 fallback으로 등록 상태를 유지합니다.

McDonald's China 공식 MCP(mcp.mcd.cn)를 DSH로 가져옵니다. 직접 생성한 token을 붙여 넣으면 handshake 후 tools/list를 호출하고 모든 원격 도구를 `mcd_` 접두사의 네이티브 DSH 도구로 등록합니다. 상류 도구 추가/삭제는 코드 변경 없이 재시작만 하면 됩니다. 상시 도구 4개(mcd_status, mcd_test, mcd_tools, mcd_call)와 원클릭 연결 테스트 설정 페이지를 제공합니다. token은 이 기기의 `~/.dsh/dsh-mcd-mcp.json`(mode 600)에만 저장하며 저장소나 대화 컨텍스트에는 들어가지 않습니다.

DSH에서 실제 ChatGPT/Codex 구독 한도를 확인합니다. `/wham/usage`에서 5시간/주간 창, 초기화 카운트다운, 추가 사용 잔액, 추가 한도 그룹, 코드 리뷰 한도를 읽습니다. 서버 발급 초기화 크레딧을 안정적이고 재시도 안전한 idempotency key로 조회/사용하고 여러 계정을 전환합니다. `/codex/models`의 실제 모델 목록을 읽으며 가져올 수 없으면 내장 목록 대신 비어 있게 표시합니다. Settings에서 브라우저 인증 또는 device code로 원클릭 로그인(콜백 붙여넣기 대안 포함); 자격 증명은 DSH credential service에 둡니다. 모델 호출은 내장 `llm-pi-ai` Codex route를 재사용하며 provider route를 등록하거나 점유하지 않습니다. 다른 플러그인이 해당 route를 소유하면 quota console만 제공합니다. 남은 한도 칩을 composer 옆에 둡니다.

입력란에서 다음에 쓸 내용을 예측해 커서에 고스트 텍스트로 보여 줍니다. 편집기 자체 글꼴과 줄 상자에 맞추며 Tab은 삽입, Esc는 무시합니다. 초안과 최근 대화 몇 턴을 넣어 세션 자체 모델 라우터를 한 번 호출합니다. 빈 입력란에도 사용자 말투의 다음 프롬프트를 제안하고 어시스턴트 반복 제안은 버립니다. Host는 loopback 전용 하위 경로 HTTP 라우트(본문 64 KB 제한, 결과 2분/캐시 미스 20초, 세션당 동시 호출 1개)를 둡니다. 클라이언트는 `conversation.input.overlay`의 높이 0 오버레이이며 `draftRev` CAS와 `setDraft` 대체로 삽입을 보호합니다. IME 조합과 slash 명령은 제외. `node:test` 16개, 런타임 종속성 없음, 실제 스크린샷은 docs에 있습니다.

에이전트가 DSH 오른쪽 사이드바 내장 브라우저를 제어하는 도구 38개입니다. 열기/이동, 페이지 텍스트·대화형 요소·접근성 트리 읽기, 클릭/입력/선택/체크/키/스크롤, 모서리 드래그로 프레임 크기 조절(좁은 프레임에서는 페이지가 반응형으로 재배치되어 모바일 레이아웃이 표시됩니다), 스크린샷/전체 페이지 캡처/PDF/Markdown/JSON 추출/다운로드/cookie, 단계별 기록 재생 및 동작 허용 목록을 지원합니다. 내장 `web_search`는 SSH 터널의 자체 호스팅 SearXNG를 제공하고 Tavily와 실제 브라우저로 대체합니다. Fetch 공급자는 차단된 GitHub URL을 측정된 미러로 재시도하고 응답 헤더 charset을 감지해 GBK/GB2312 페이지를 올바르게 디코딩합니다.

전체 너비 DSH 페이지에서 ChatGPT Dots를 사용합니다. 데스크톱 시작 시 미리 로드하고 같은 DSH 세션에서 패널을 전환해도 페이지를 유지하며, 중/영 테마 대응 제어, 새로고침, 외부 브라우저 대체를 제공합니다. Windows DSH 0.2.1-rc.3용 로그인 유지 컴패니언은 별도 적용합니다. DSH 0.2.1-rc.5로 업그레이드한 후 재시작 시 내장 Dots 페이지에서 다시 로그인해야 할 수 있습니다. 현재 호스트 브라우저는 재시작 간 로그인을 유지하지 않으며 호스트 업그레이드가 이전 추가 유지 처리를 덮어쓸 수 있습니다. 이번 Dots 업데이트가 로그인 저장 기능을 제거한 것은 아닙니다.

DSH의 프로세스 전체 프록시 경로로 동작하는 loopback relay입니다. TUN fake-ip에서도 `web_fetch`가 작동합니다. 프록시가 죽으면 같은 요청에서 직접 연결로 대체하고 probe가 성공하면 다시 프록시로 돌아갑니다.

선택적으로 Decision API 도구를 검색하고 검색 소스를 미리 가져옵니다. 세션별 제어와 일반 검색 방식으로의 대체를 지원합니다.

모델 장애 조치 대기열입니다. P1 할당량 초과, 요청 제한, 사용 불가 시 대기열의 다음 모델로 자동 전환해 계속 처리합니다. 모든 설정은 GUI에서 할 수 있습니다.

DeepSeek Harness용 하이브리드 model router입니다. 선언형 규칙, 사용자가 설정하는 비용 제어(cost-first/quality-first/balanced), 선택적 LLM task classifier, 실패 시 fallback chain을 제공합니다.