Plugins

DeepSeek Harness (dsh) plugin: live local LLM endpoint health in the web rightbar. llama.cpp slots (busy/TTFT/wedged, /metrics) + Ollama MODELS (loaded vs empty). Engine fingerprinted by response shape. Installable dsh.bundle.

Model relay station for DeepSeek Harness (模型中转站): serve any registered DSH LLM provider (e.g. @tnnevol/dsh-codebuddy) over an OpenAI-compatible /v1 API, so other projects can call those models with a standard OpenAI client.

DeepSeek Harness (dsh) plugin that connects Sber GigaChat models: a built-in local OAuth proxy (client-credentials -> OpenAI-compatible endpoint, upstream serialization, legacy tool-calling translation) plus automatic provider-route bootstrap, so GigaChat

A searchable, sortable model picker for the DeepSeek Harness Web UI — provider groups, favorites, hidden-models blacklist, live per-task cost, prices and context windows, a refresh button that re-syncs every provider model list from its live API, and the

Keep DSH provider model catalogs current with the living models.dev catalog: refreshes each provider's models (new and existing) with updated modalities and thinking levels — automatically and from one refresh button in the Models page. / Utrzymuj żywy ka

Measure whether a dsh setup change actually helped: register repeatable cases, run them, and diff the results before and after you change rules, skills, or models.

Declares the OpenCode Go and Command Code route wire protocol and endpoint (so the API address never has to be typed), adds the DeepSeek V4.1 models to the OpenCode Go route and puts them first, completes the Command Code model list from its public catalog, and attaches a per-conversation x-opencode-session header to OpenCode requests (fixes 400 MissingSessionID).

Enable or disable model providers, search models in the selector, and rename provider display names.

OpenRouter provider that reads the live /models list, so new models appear without a harness release: models sorted by name with input/output prices, per-model reasoning levels, image input, reasoning_details replay, Anthropic prompt caching, and actual spend per session including subagents from OpenRouter usage.cost, shown under the Web UI composer and by /cost.

OpenCode Go provider that verifies a model against the gateway rather than treating a recommendation as a measurement: the models.dev protocol suggestion is probed before it is used, reasoning levels follow a measured contract, and requests run through an ordered protocol fallback chain. Context window, max output and image-input numbers are declared by a runtime-refreshed models.dev layer rather than probed, with operator overrides above both. Several OpenCode Go credentials can be stored and switched from the settings page.

Brings the Codex (ChatGPT Plus/Pro) subscription already signed in on this machine into DSH. It (1) shows the 5-hour and weekly windows with their reset times and the remaining reset credits in a sidebar panel, with one-click reset that consumes a credit; (2) registers the subscription route in llm-pi-ai (openai-codex) when the quota is read, so the Codex models appear in the model picker; (3) refreshes the OAuth access token before expiry and writes it back atomically, so no key has to be replaced by hand; and (4) adds a codex tool that delegates a task to the local Codex agent. No API key has to be entered.

Discovers OpenRouter's free models, measures their real speed and context size, and writes them into the DeepSeek Harness model list fastest-first, with speed and context in each display name, plus a Settings section to configure and run it.

Declares the wire protocol for the opencode-go route so models the installed catalog does not describe resolve, adds the DeepSeek V4.1 models to that route when they are missing, and attaches a per-conversation x-opencode-session header to OpenCode requests.

AI-native expressions for two-way chat: people see images, while text-only models read fixed meanings; includes a picker, custom expressions, import/export, and two visual styles.

Speaks your agent's reports in a voice you choose, compressing long reports before speaking. Character voices need no training - a 3-10 second reference clip clones one, and community-trained models work too - and no 6.4 GB GPT-SoVITS install: the plugin installs its own runtime and voice. An existing GPT-SoVITS can be used instead, and it is the same voice model.

Use your DSH models in [Libre WebUI](https://github.com/libre-webui/libre-webui) for chat, sandboxed Work and usage tracking, with provider credentials kept in DSH. Requires Libre WebUI 0.37.0 or later.

Replaces the chat input model picker with a provider-grouped, collapsible list: groups models by provider, folds vision-bridge mirror providers back into their upstream, and carries the reasoning-effort options each model declares in settings. Tested on DSH Desktop only; the DSH web build is untested.

AlphaRSI — three-role self-improvement loop that evolves the training recipe of alpha sequence models. Operator runs walk-forward, Evaluator grades with OOS RankIC, Refiner proposes hyperparameters via TPE Bayesian search; only strictly-better recipes are kept. Offline one-command run, data bundle shipped with sha1 reconciliation, GPU auto-detect. Ships one skill.

Mirrors cc-switch providers and models into DeepSeek Harness the moment they are saved, and shows each provider's plan windows, account balance, or custom usage script in the composer and the settings card.

Keeps a redundant sandbox-escalation argument from failing a tool call. Escalating tools (pwsh, bash, write, edit) advertise the full sandbox_permissions enum, but DSH only accepts a level strictly wider than the one already in effect — a check its own source calls 'deliberately not a schema constraint'. A model that reflexively passes the argument therefore picks the level it is already at and the call dies before running with 'sandbox escalation to "workspace-write" is not strictly wider than this call's current "workspace-write" mode', costing some models a whole turn of retries. This plugin registers one tools/execute waterfall listener and, only on that documented rejection and only when the arguments really carried an escalation field, re-issues the identical call once without it. Safe by DSH's own documentation: the rejection precedes any execution ('nothing has run'), and the corrected call cannot match again, so the retry is loop-free. Reproduced and verified end to end — before: isError true, the command never ran; after: the command's real output, isError false, one tool/call and one tool/result. Zero dependencies.

Large-image recognition for vision-exp models: lossless 800×800 tile recognition (smart/pipeline/full), local OCR with preprocessing & handwriting routing, optional multi-vendor GPU (DirectML/CUDA/OpenVINO) with auto CPU fallback.

Codex-style DeepSeek Harness model and reasoning selector with model-advertised effort levels, DSH-native themes, left-clipped radiation effects, and copy-ready reasoning-effort declaration guidance for custom-provider models.

A Models settings companion page listing every model's input modalities in one place, editable for both LLM adapter families

A dsh plugin that replaces the built-in llm-pi-ai and llm-deepseek adapters, the model selector and the Models settings page with self-maintained versions, tracks upstream @earendil-works/pi-ai, and adds quota/balance lookups plus a provider management UI