Plugins

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.

Servicio HTTP/SSE loopback para que un plugin de Obsidian controle sesiones DSH (listar, crear, enviar) e inyecte hasta 10 notas de la bóveda en el system prompt. Necesita el plugin Bridge complementario activo dentro de Obsidian.

Plugin de activación por eventos (el timbre junto al despertador de dsh-proactive): fs watchers recursivos con debounce y filtros glob por regla; endpoint webhook solo loopback con token por regla; reglas que despiertan una sesión objetivo mediante el servicio agents del host (follow-up directo si está activa, reanudación previa si está inactiva) e inyectan un prompt con variables de plantilla paths/count/payload. Recarga en caliente el archivo de reglas sin reiniciar DSH y añade cada evento a un historial JSONL auditable, consultable con trigger_list y trigger_history. Una ruta health loopback informa del estado de instalación y ejecución. Usos: ordenar descargas, diagnosticar alertas CI/monitorización de scripts locales y vigilar logs. JavaScript puro, sin compilación, MIT.

Gateway de emparejamiento para DeepSeek Harness. Por defecto escucha HTTP sin cifrar en `127.0.0.1:3443` y hace proxy inverso al DSH local `http://127.0.0.1:3080`, sin certificado para instalaciones solo loopback. Producción, acceso entre máquinas y emparejamiento de apps requieren TLS: configura `tlsCert`/`tlsKey` y `publicUrl` HTTPS (se admite autofirmado). Caddy/Nginx frontal puede terminar TLS público, pero el gate también necesita esos archivos y el proxy debe volver por HTTPS. Incluye QR con TOTP, usuario/contraseña, sesiones deslizantes con límites absolutos, revocación de dispositivos, panel admin Web/log de auditoría y túnel WebSocket. Asistente Web para la primera configuración. Pruebas declaradas: 171/171 unitarias, e2e de 29 pasos.

Notificaciones de escritorio y distintivo de no leídos en el icono para DeepSeek Harness. Avisa al terminar un turno, si espera aprobación/pregunta/confirmación de plan o si falla. El distintivo cuenta Sessions que requieren atención y se borra al verlas; al hacer clic en un banner, cambia a esa Session. Sigue el idioma DSH; subagent silenciosos por defecto; título y longitud del cuerpo configurables; interfaz HTTP loopback para scripts y herramientas.

Reorganiza el transcript de una Session de DeepSeek Harness en un documento Markdown estructurado con esquema e índice, no como exportación cronológica del chat. Disponible como herramienta model `session_to_doc`, botón composer y endpoint Web solo loopback.

Ejecuta el CLI local Codex desde DSH con cuatro herramientas (codex_run, codex_resume, codex_queue, codex_probe) e incluye un bridge loopback independiente con API HTTP y mailbox de archivos.

Ejecuta el CLI local Claude Code desde DSH con tres herramientas (claude_run, claude_resume, claude_probe), analiza `--output-format json` y devuelve el session id para continuar; incluye un bridge loopback independiente.

Accede a este Harness desde el teléfono por una red privada: gateway que conserva loopback, códigos de emparejamiento de un solo uso y diseño vertical para pantallas estrechas.

Auditoría de solo lectura para comprobar si las rutas HTTP de un plugin DSH están realmente protegidas. Descubre las rutas registradas leyendo el código del plugin y solo prueba loopback.

Guía de despliegue de NapCat en Windows y comprobaciones de solo lectura del puerto loopback para DSH, con un lanzador de administrador iniciado por el usuario.

Predice lo que vas a escribir y muestra texto fantasma en el cursor, medido con la fuente y la caja de línea del editor; Tab lo inserta y Esc lo descarta. Hace una llamada breve al router de modelos de la sesión, con el borrador y los últimos turnos; también sugiere el siguiente prompt con la voz del usuario si el campo de entrada está vacío y descarta repeticiones del asistente. Host: ruta HTTP loopback con subrutas, cuerpo máx. 64 KB, caché de resultados 2 min y de fallos de caché 20 s, una llamada simultánea por sesión. Cliente: overlay de altura 0 en `conversation.input.overlay`; protege la inserción con CAS `draftRev` y fallback `setDraft`. Omite composición IME y comandos slash. 16 pruebas `node:test`, cero dependencias de ejecución, capturas reales en docs.

Enrutamiento OpenRouter por sesión: un relay loopback escribe `service_tier`, `provider.sort` y la variante `:floor` en cada solicitud según uno de nueve presets elegidos en el chip del compositor. Cada preset fija una URL base para valores predeterminados por ruta.

Relay loopback como ruta de proxy de todo el proceso DSH: `web_fetch` funciona con TUN fake-ip; si el proxy cae, la misma solicitud pasa directamente y una prueba vuelve a activar el proxy.

Controla un cliente local MaaEnd (MXU) mediante su API HTTP loopback: 10 herramientas de agente `maend_*`, un panel lateral y pasos de silencio al iniciar/cierre del juego condicionado al éxito en la cadena de tareas. Requiere instalar MaaEnd localmente y tener activo el servicio MXU.

Pool de rutas salientes DSH: 15 seeds + 20 candidatas, 7 tipos. Verifica antes de usar con handshake TLS y cadena completa (TCP solo puede dar por sana una IP muerta); cooldown tras fallos 2/5/15 min y ruta sana más rápida por tipo. Reescribe URL con 4 reglas (prefix/registry/index/host-replace) para git/pip/npm/pnpm/gh/curl/wget por shell bash (solo descargas que ejecuta el modelo; no pwsh/host/renderer), tools line_pool_status/line_pool_run o child-process spawn. Conserva saltos de línea y cambia solo el segmento download. Nunca reescribe user:token@host; no envía tokens a mirrors ni modifica config global. Panel: latencia/error/cooldown, retest/activar por ruta, estado TUN/proxy mediante network-adapter y loopback. Revalida candidatas cada 24 h.

Previsualiza un OpenWiki del workspace en la barra lateral derecha de DSH Web: detecta el directorio wiki, lo sirve con el lector oficial `openwiki visualize` en un puerto loopback y lo integra en su propia pestaña. Compara las versiones instaladas de openwiki y DSH con mínimos configurables y consulta el registry para comprobar sus propias actualizaciones.

Conecta un inicio de sesión local existente de MiniMax Code con DeepSeek Harness mediante un proxy inverso loopback, sin guardar una API key ni cerrar sesión.

Expone todos los modelos de todos los proveedores configurados en DeepSeek Harness mediante un único endpoint compatible con OpenAI en un puerto loopback fijo. Una página lateral muestra base URL, key, puerto e IDs de modelo.

Convierte la barra de navegación entre turnos de DSH en un visualizador musical: cada barra fina crece y se ensancha según su banda de frecuencia, usando audio del sistema capturado mediante WASAPI loopback. Solo cambia la geometría, sin color ni brillo; conserva la vista previa al pasar el cursor, el salto al hacer clic y el desplazamiento con la rueda.

Reimplementa en loopback la API de cartas Olivia Lin de BSide, ya cerrada, para que el cliente del juego siga enviando y recibiendo. Cada carta entrante llega a una sesión DSH persistente y la respuesta vuelve al almacén local.

Control del gateway LAN dsh-lan-gate: botón junto a Settings muestra si funciona y cuántos teléfonos/tabletas esperan aprobación; abre la página de aprobación limitada a loopback sin escribir la URL. Si está detenido, inicia primero la tarea programada. No aparece en superficies móviles.

Restaura ajustes persistidos en el host cuando dsh web se abre tras un proxy inverso con un nombre de host no loopback. Marca la conexión de página como propiedad del host para cargar Settings y Models en vez de indicar que los ajustes no están disponibles en este navegador. Para dsh 0.1.7–0.2.x.

Transmite por loopback el estado del agente de DeepSeek Harness (estado, llamadas a herramientas, errores y esperas de aprobación) a WhalePal, una app independiente de mascota de escritorio, para que siga el estado real del agente y no lo deduzca de capturas.