Plugins

Unified model routing for DeepSeek Harness: one logical ModelID over multiple providers with first-token failover and cooldown, health-aware candidate ranking, three tiers (tier1/tier2/tier3) auto-selected by purpose, per-candidate reasoning effort, and a settings-page management panel.

Daily cost card for your local LLM gateway: /usage reads gateway_usage.jsonl and itemizes spend per model and per hour, covering official direct, OpenRouter relay, and in-plan metered pools.

Automatic Flash/Pro routing for DeepSeek Harness subagent calls: light tasks run on a cheap model (e.g. V4 Flash), heavy tasks stay on the strong model — with per-call overrides and /delegate session modes.

Composer-dock balance/usage readout for DeepSeek, Kimi Coding, OpenCode Go, Zhipu, MiniMax, OpenRouter and OpenAI Codex — provider-based, 2s polling, 5min cache.

A model-native meta-agent plugin for DeepSeek Harness: the model synthesizes task-specific workflows at runtime from five agentic workflow patterns (prompt chaining, parallel workers, router, supervisor, evaluation loop) and coordinates tools and subagent

Integrated control plane for the local strata model stack: manage the strata-hop :8080 model-switching router (status/models/warm/logs), stream live model-switch and load state into the web GUI status chip and the strata_status tool, preload a model engine the moment it is selected, edit run-config and shared chat settings from a GUI tuning panel or strata_params, and ship the dsh provider patch that routes chats to the router with Effort-level mappings. Read-only toward inference: it never sends a prompt and spends no tokens; bring your own running strata plus strata-hop (defaults :8080 hop, :8756 monitor). As shipped, the status server listens on 0.0.0.0:8756 with no authentication and open CORS, and its endpoints can change run-config and restart the engine; set host to 127.0.0.1 in cordis.patch.yml to keep it on this machine. It expects strata-hop as a systemd user unit.

Manually resets old AgentRouter Responses replay state for chats blocked by upstream resource-binding errors while preserving history; supports DSH 0.2.2-rc.1.

OpenAI-compatible gateway for every connected model: groups with failover, aliases, session stickiness, usage charts, API keys and a LAN listener (open to the whole network until you create an API key).

Provider limits in the DSH sidebar footer: the remaining OpenRouter credit balance and OpenCode Go rolling 5-hour, weekly and monthly quota windows, shown as bars or percentages with a Remaining/Used choice, plus a Creator-mode request when the active provider is not supported.

Keeps prompt caches warm for OpenCode Go DeepSeek conversations by predicting each request's cache hit before it is sent, rotating the x-opencode-session routing id when that request would be cold and recent requests are slow, and pooling idle warm ids for reuse by new conversations.

System Duo for DeepSeek Harness. Make your agent think fast and slow: System One decision models triage, guard, shield, mask and check your agent loop beside System Two (vllm-sr/auto), with a free app.vllm-sr.ai key.

Model failover router that queues every connected DSH model into priority templates, retries a failing candidate before switching, cools down rate limits briefly, quarantines failures per session, and auto-sends a continue message when a turn fails or hits the token ceiling.

Routes subagent delegations once at spawn time by keyword rules over the delegation label and prompt: the first matching rule sets the child's provider/model/reasoningEffort, explicit caller routes are kept unless a rule opts in, and unmatched delegations pass through unchanged.

Deep-sea tidal-blue theme for the DeepSeek Harness Web client with a single accent color and matte (non-glass) surfaces in light and dark modes. Same-token UI additions: a single-line dock HUD reading real session projections (context usage vs. window, TPS, cache-hit rate, elapsed time), per-turn subagent delegation cards showing token usage and an optional cost estimate from OpenRouter public prices, sidebar pagination with pinned sections and type-to-filter, and model / reasoning-effort / permission controls under the composer. Each addition can be toggled in the plugin settings.

Xray-core supervisor and router for DSH with VLESS import, global/China-direct/whitelist routing, a three-card live panel, and self-update.

Transparent cross-provider failover for DSH Agent main requests: named fallback chains, per-session chain choice, per-route retry overrides.

Qoder supplier plugin for dsh-router, covering both the CN and international regions. Uses a personal access token for authentication, signs requests to the Qoder cosy gateway and converts upstream SSE to OpenAI SSE; the model list reports the upstream display name alongside the key, while account selection and cooldown stay in dsh-router core.

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/.

DSH Web 插件：实时显示模型供应商额度与用量（DeepSeek / GLM Coding Plan / OpenRouter / Kimi Coding Plan + 自定义适配器）

Router and artifact gate for DeepSeek Harness: a per-turn skill router that ranks the live catalog, plus a write gate enforcing read-before-write, dedupe and a four-line audit on every file write.

Hides tool schemas from the model's prefill and reveals one tool at a time through a decision model API. Requires a running decision-model API endpoint and pnpm on PATH.

Two-level model routing: pick the best model group first, then the best channel within that model; a failure switches channel before switching model. Preference mode pins one model across its channels, a status bar under the composer shows the live model, channel health and the day's token usage, and a speed/intelligence report can be imported as the routing prior. Exposes a model_orchestrator tool and a /route command.

Adds the Liquid Inference auction exchange as a model route, with liquid.auto and its ranking variants, and four tools that read offers, markets, routers and job receipts. Requires a Liquid Inference API key (LIQUID_API_KEY, spend scope).

Derivative of dingminhua/dsh-subagent-default-model (do not install both): routes subagent delegations by role with candidate pools, cross-provider failover and fork window slicing; shows running subagents in a header panel with duration and token counts; supports mid-run and batch model switching and exposes routing tools the model can call. Subagent pages also carry a context / usage stats strip (context occupancy, output speed, cache hit rate, token buckets) and a model menu limited to that page’s whitelist — a pick outside the whitelist is substituted with an allowed model instead of running unlisted.