124 results
DSH plugin: manage local model inference servers (llama.cpp / vLLM / SGLang) — registry, safe start/stop, parameter profile versioning with validation. Red lines: never pkill, never stop protected port 11437, never auto-restart dsh.
DeepSeek Harness plugin: fingerprint the provider, model id, sampling, offered tool schemas, and system-prompt prefix of each request, and diff the live card against a pinned one through the modelprint_check tool
DSH web UI 插件：把输入框官方「模型选择」控件替换为两个独立按钮（供应商选择 + 模型选择，两级联动），保留推理等级调节；只动这一处控件，不改 DSH 源码。
DSH client plugin — keyword search when selecting models in the Web GUI
Aggregate DSH plugin bundle — one install, five plugins: dsh-memarc (long-term memory + session archive), dsh-session-git (export/import sessions as committed JSON), dsh-session-branches (session fork tree), dsh-goal-auto-resume (re-arm unfinished goals), dsh-model-reasoning (per-model reasoning capability & effort editor). Declares dsh.bundle, so the whole repository installs as a single profile layer: no clone, no build.
DeepSeek Harness model selector upgrade: one-level model & reasoning-effort picker with 5 Liang tiers, custom model aliases, and peak/off-peak price countdown.
dsh web 插件：自动配置模型思考档位与能力，并按 provider+model 记住最后选择的思考档位
Filter input for the composer model menu: case-insensitive list filtering with combobox keyboard navigation, rendered with the shipped menu design
Persistent DSH model-account authorization UI for ChatGPT, Claude, and other llm-pi-ai login flows
Structured cross-model peer review for DeepSeek Harness
A DeepSeek Harness plugin with a vertical model list and an animated Three.js particle reasoning-effort control.
Automatic per-task model hot-switching for DeepSeek Harness (dsh): image-aware tasks route to the vision model automatically, every other task keeps your default model. Zero extra tokens, no context disturbance.
DSH model channel manager: roundrobin failover engine (host) + model config view tab (client).
mycel working modes: a switch_mode tool returning one selected mindset as an ordinary tool result, plus a shared core-behavior system-prompt section
Automatic reasoning and image capability detection for custom DeepSeek Harness models
DSH session model selector with search and reasoning-effort pane: shadows the native conversation.input.model seat (independent cordis client plugin).
DSH plugin: declare VIRTUAL model ids bound to a pluggable routing algorithm (priority, round-robin) over real provider/model candidates; using the virtual id transparently dispatches to a real provider chosen by the algorithm, with failover.
Model availability tester panel in the Models settings page: one-click real streaming test per added model with TPS / first-token / elapsed metrics
DeepSeek Harness plugin: keep every model request inside an allowed set per provider, redirect or refuse the rest, and warn when a request would make a llama.cpp router reload models
Declare whether a configured third-party model accepts image (multimodal) input; writes the modality into the owning provider settings and verifies it through runtime model resolution.
Platform and manufacturer grouped model picker for DeepSeek Harness
DSH compatibility guard: repairs compaction limits, fail-safe sandbox escalation, and missing tool descriptions for third-party models; capability lookups are cached and the deprecated prompt-injection path is off by default.
DSH web plugin: searchable model selector for DeepSeek Harness WebUI
Model router for DeepSeek Harness: keep a pool of models and every step is routed to one of them with its reasoning effort set — a click-only composer panel, three authorization scopes (a manual pick only counts in the scope that owns it, and effort scope is bound to the one model you choose), task presets with per-model weights, vision routing for image steps, session signals (todo, context pressure, token budget, subagents) and an optional semantic task classifier. Thinking stays on, because DeepSeek rejects a tool call made with thinking off.