1090 results
Agent2Agent interop for the DeepSeek Harness in one package — the realtime mesh hub (project registry, immutable message history, WebSocket presence) and mesh client (per-agent connections, serial injection), the model-facing a2a_peers / a2a_message / a2a_history tools, and the /a2a command surface
Generate and display images inline in the DSH chat via API channels or local CLIs (mmx / codex / agy), and read images into structured JSON evidence (OCR / layout / semantics) on any model — with backend probing and a bundled recovery skill.
Windows-first bash tools for DeepSeek Harness: git_bash (Git Bash) and wsl_bash (WSL2) model tools with bwrap sandbox, approval mode, terminal-style UI cards, and background-job support
DSH plugin exposing a single `sleep` tool that lets the model pause for a configurable number of milliseconds, honoring cancellation.
Unified DeepSeek Harness plugin: role-based subagent routing + per-agent evolution — prefercmd/memory as knowledge allow/deny lists, so repeated tasks start from proven commands and save tokens
Codex App Server model provider plugin for DeepSeek Harness
Model-selectable subagents and guided model routing setup for DeepSeek Harness
Tavily for DeepSeek Harness: opt-in advanced Tavily model tools (search, extract, map, crawl), multi-key management, usage gauge, and the Tavily Search settings card. The built-in web_search is never replaced.
dsh plugin: recognize attached images with the built-in Windows OCR engine (Windows.Media.Ocr) and send only the recognized text to the model. Text models never receive image bytes; vision passthrough is opt-in.
Reasoning effort settings and per-call subagent model routing for DeepSeek Harness custom providers
NoLetMe — real-time reasoning-trajectory keyword stats for the DeepSeek Harness web chat. Tracks the We-need/Let's efficient trajectory vs the Let-me hesitant trajectory vs the The-user-wants neutral framing in the model's streaming reasoning blocks.
Route DeepSeek Harness agent requests by role: planner model in plan mode, default model otherwise, subagent model for subagents.
dsh client bundle: a sidebar usage & cost dashboard — token stats, peak/off-peak aware per-model pricing, membership ladder, achievements and a daily heatmap.
DeepSeek Harness web plugin: DeepSeek API spend (per session / all sessions) and account balance in a top-right overlay card. Only deepseek-official routes are billed; third-party models are reported as unbilled.
Model-based permission approval (approve-for-me) for DeepSeek Harness: an approval/request answerer backed by a separate reviewer model
CC Switch-style token consumption statistics for the DeepSeek Harness Web GUI: per-request usage log, real-consumption hero, request-time-bucketed trend chart, per-model peak/off-peak list pricing, per-project and per-model stats, and account balance.
Eyes for text-only DeepSeek on DeepSeek Harness: image understanding via Doubao Web by default (zero cost, no API key), Antigravity IDE quota (flash/pro), Gemini API, or Cockpit proxy — model-invokable vision tool, wrapper adapters, evidence memory, and a bilingual client panel.
A searchable, sortable model picker for the DeepSeek Harness Web UI — provider groups, favorites, live per-task cost, prices and context windows.
Inspect and manage the live instruction chain and project skills for the current dsh session, with in-place creation and editing — what you see is what the model gets
DeepSeek Harness plugin: a sidebar "today's usage" card that pops a token usage dashboard (today / 7 / 30 days, 30-day per-model stacked chart, model ranking with usage share, conversation stats). Data comes from the session corpus only (idempotent rebuild, no double counting), persisted across restarts.
Rewind a DSH conversation to any user message: hide the tail from the chat view and the model context, keep the durable log intact, cancel freely before sending.
Free vision plugin for DeepSeek Harness (dsh): image understanding for text-only models with free-tier providers (Qwen3-VL-Flash / DeepSeek-OCR / Doubao). 免费视觉插件：纯文本模型看图能力，优先免费模型（通义千问 / 硅基流动 / 豆包）。
GateCraft（门控工艺）：门控式数学建模 skill 套件 for DeepSeek Harness——五阶段流水线、约束分级（事实/不变量/观察）、国奖写作范式与 docgate 机械验收；agent 求解质检，人在阶段门思考决策。
Brings open-mcp-apps into DeepSeek Harness (dsh): an Apps section in the sidebar, one container per app with its own workspace and conversation, an agent-presence strip under the app, and inline app rendering for MCP tool calls.