Deeptop

English | 中文

Deeptop is a lightweight native desktop client for DeepSeek Harness. It is not a wrapper around dsh web: the Deeptop bridge is itself a Cordis profile bundle, and the Agent, session store, model routes, tools, presets, skills and workspace services stay inside the same DSH tree.

Project: Sparrived/DSH-Deeptop

Current surface

The desktop workbench currently provides the core WebUI conversation surface:

• persistent session list, history recovery and event-driven live updates;
• lazy new-session creation, session rename, fork, search and queue removal;
• model catalog and per-session model selection;
• optional native workspace selection and workspace registration;
• queue/steer prompt modes, stop current turn, tool-call/result rows and the turn-aware trajectory ledger;
• native approval and question response paths;
• native DSH workbench with a Profile roster, Skill catalog, Subagent history/follow-up/interruption, Goal lifecycle, Host settings, and Provider/model catalogs;
• runtime inspector for DSH host, Cordis profile, workspaces and active routes.

The bridge forwards the same DSH ApiProxy domains used by the WebUI: sessions, subagents, skills, goals, settings, credentials, provider discovery, directory browsing, workspaces and preset authoring. The native UI consumes those domains directly; it does not duplicate plugin logic in the desktop process. If an optional domain is absent from a profile, its panel stays unavailable while the Agent conversation remains usable.

Runtime model

The desktop process starts one hidden, long-lived DSH process:

Tauri native window + JSONL stdio
  -> npx @deepseek-ai/dsh@latest --profile desktop
  -> dsh-base + deeptop-bridge + user desktop profile bundles
      -> Cordis services, Agent presets, sessions, tools and ApiProxy

The application materializes $DSH_HOME/profiles/desktop on first start and preserves user-added profile bundles. The bridge package is written to $DSH_HOME/profiles/node_modules/deeptop-bridge, so the current npx @deepseek-ai/dsh@latest package can resolve it without a separate global installation.

The DSH process uses $DSH_HOME/desktop-runtime as its default current directory. A selected workspace is passed to session.create({ cwd }), so each session can own its own working directory without making DSH depend on the desktop application's project directory.

Development

Requirements:

• Node.js 22.19+ or 24+;
• Rust/Cargo for Tauri;
• npx available on PATH;
• network access to the configured npm registry on first DSH use.

npm install
npm run tauri:dev

For a frontend-only preview:

npm run dev

The runtime intentionally follows the current @deepseek-ai/dsh@latest package. User DSH configuration and profile bundles remain under the configured DSH_HOME.