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.

Loopback HTTP/SSE service that lets an Obsidian plugin drive DSH sessions (list, create, send) and inject up to ten vault notes into the system prompt; needs the companion Bridge plugin running inside Obsidian.

Event-driven trigger plugin (the doorbell beside dsh-proactive's alarm clock): recursive fs watchers with per-rule debounce windows and glob filters, plus a loopback-only local webhook endpoint with per-rule tokens, fire rules that wake a target session through the host agents service - live sessions get a direct follow-up, cold ones are resumed first - and inject a rendered prompt with template variables (paths, count, payload). Rules hot-reload from the event-gate rules file without restarting DSH, every trigger appends to an auditable JSONL history queryable via the trigger_list and trigger_history tools, and a loopback health route reports install and runtime status. Typical uses: auto-organizing a downloads folder, diagnosing CI or monitoring alarms posted from local scripts, and watching logs for errors. Pure JS, no build step, MIT.

Device-pairing gateway for DeepSeek Harness: by default it listens on 127.0.0.1:3443 over plain HTTP and reverse-proxies the local dsh at http://127.0.0.1:3080, so a loopback-only install works without any certificate. Production, cross-machine and app pairing require TLS: configure tlsCert/tlsKey plus an https publicUrl (self-signed is fine). A fronting Caddy/Nginx may terminate the public TLS, but the gate itself still needs tlsCert/tlsKey and the proxy must connect back over HTTPS. Features TOTP-authorised QR device pairing, username/password login, sliding sessions with absolute caps, device revocation, a WebUI admin panel and audit log, and a reverse proxy with WebSocket tunnelling. First-run setup is completed in the WebUI wizard. Tests: 171/171 unit tests, 29-step e2e.

Desktop notifications and an app-icon unread badge for DeepSeek Harness. A banner fires when a turn finishes, when an approval, question or plan confirmation is waiting, and when a turn fails; the badge shows how many sessions need you and clears once viewed, and clicking a banner switches to that session. Follows the DSH interface language, subagent sessions stay quiet by default, titles and body length are configurable, and a loopback HTTP interface lets scripts and other tools post notifications.

Reorganizes a DeepSeek Harness session transcript into a structured Markdown document with an outline and a table of contents instead of exporting the chat in chronological order. Exposed three ways: a model tool session_to_doc, a composer button, and a loopback-only Web endpoint.

Runs the local Codex CLI from DSH as four tools (codex_run, codex_resume, codex_queue, codex_probe), and ships a standalone loopback bridge with an HTTP API and a file mailbox.

Runs the local Claude Code CLI from DSH as three tools (claude_run, claude_resume, claude_probe) that parse --output-format json and return the session id for follow-ups, plus a standalone loopback bridge.

Reach this Harness from your phone over a private network: a loopback-preserving gateway with one-time pairing codes, plus a portrait layout for narrow screens.

Read-only audit of whether a DSH plugin's HTTP routes are actually fenced, discovering the registered routes by reading plugin code and probing loopback only.

Windows NapCat deployment guide and read-only loopback port checks for DSH, with a user-initiated administrator launcher.

The composer predicts what you are about to type. Inline ghost text sits on the caret, measured against the editor's own font and line box, and Tab turns it into real text (Esc ignores it). The prediction is one short call through the session's own model router, seeded with your draft and the last few turns of the conversation, so an empty composer also gets a next-prompt guess written in the user's voice; suggestions that merely restate the assistant are dropped. Host side is one HTTP route with sub-path endpoints (loopback-only, 64 KB body cap, 2-minute result cache, 20-second miss cache, one in-flight call per session); client side is one 0-height overlay registered on conversation.input.overlay, with insertion guarded by the composer's draftRev CAS and a setDraft fallback. IME composition and slash commands are skipped. 16 node:test cases, zero runtime dependencies, real screenshots in docs/.

Per-Session OpenRouter routing: a loopback relay writes `service_tier`, `provider.sort` and the `:floor` variant into each request from one of nine presets picked on the composer chip, with a pinned base URL per preset for route-level defaults.

Loopback relay as DSH's process-wide proxy route: web_fetch works under TUN fake-ip; a dead proxy gets same-request direct fallback, and a probe switches back.

Drives a local MaaEnd (MXU) client over its loopback HTTP API: 10 maend_* agent tools, a sidebar dashboard panel, and startup-mute / success-gated close-game steps in the task chain. Requires a local MaaEnd install with its MXU service running.

Outbound route pool for DeepSeek Harness. Ships 15 seed routes plus a 20-entry candidate pool covering 7 target kinds (GitHub raw/codeload/main, npm registry, PyPI index, HuggingFace, general). Routes are verified before use: upstream IP lists go through a TLS handshake with full certificate-chain validation (a bare TCP check can be answered locally and report dead IPs as healthy), and failures advance a cooldown ladder of 2/5/15 minutes with the fastest healthy route picked per target kind. Routes are applied by rewriting the URL with one of four rules (prefix, registry, index, host-replace), reaching git, pip, npm, pnpm, gh, curl and wget through three paths: the shell seam (download commands the model runs through the bash tool are intercepted and rewritten automatically; commands wrapped by the pwsh tool, and fetches made by the host itself or the renderer, are not covered), two registered tools the model can call directly (line_pool_status, line_pool_run), and child-process spawn. Multi-line commands keep their line breaks and only the download segment is touched. URLs carrying credentials (user:token@host) are never rewritten, so tokens are not forwarded to third-party mirrors. No global config is modified. A sidebar panel lists every route with latency, failure reason and cooldown countdown, supports per-route retest and enable/disable, and shows live TUN/proxy status from network-adapter and loopback port checks. The candidate pool is re-verified every 24 hours so newly usable routes are added and dead ones dropped. 570 tests, 87% line coverage.

Previews a workspace OpenWiki in the DSH Web right Sidebar: detects the wiki directory, serves it with the official `openwiki visualize` reader on a loopback port, and embeds it in its own tab, reporting the installed openwiki and DSH versions against configurable floors and checking the registry for its own updates.

Connects an existing local MiniMax Code sign-in to DeepSeek Harness through a loopback reverse proxy without storing an API key or signing out.

Exposes every model of every provider configured in DeepSeek Harness through one OpenAI-compatible endpoint on a fixed loopback port, with a sidebar page for the base URL, key, port and model ids.

Turns the DSH turn-navigation rail into a music visualiser: each thin bar grows and thickens with its own frequency band, from system audio captured with WASAPI loopback. Geometry only, no colour or glow, and the rail keeps its native hover preview, click-to-jump and wheel scrolling.

Reimplements the shut-down BSide: Olivia Lin letter API on loopback so the game client keeps sending and receiving letters: every incoming letter is delivered to one persistent DSH agent session, and the reply is written back to the local letter store.

Control surface for the dsh-lan-gate LAN gateway: a button beside Settings that shows whether the gateway is running and how many phones or tablets are waiting for approval, opens the loopback-only approval page in one click instead of typing its URL, and starts the gateway through its scheduled task first when it is not running. Not rendered on phone or tablet surfaces.

Restores Host-persisted settings when dsh web is reached through a reverse proxy on a non-loopback hostname: marks the page connection as host-owning so the Settings and model pages load instead of reporting "settings are unavailable in this browser". For dsh 0.1.7 – 0.2.x.

Bridges DeepSeek Harness agent state (status, tool calls, errors, approval waits) over loopback to WhalePal, a standalone desktop pet app, so the desktop pet follows the real agent state instead of guessing from screenshots.