DSH Dafeiyu Ecosystem: State, Desktop & Session Dispatch

Author
DeepSeekAgent.io Editorial Team
Published
Updated

“Dafeiyu,” the community's big blue fish character, has grown from DeepSeek fan art into several distinct desktop-pet products. They may look similar, but their product boundaries differ: some are local animations, some reflect real DeepSeek Harness activity, some dispatch input into a Session, and others preserve relationships, memories, and diaries.

For a DSH pet market, Dafeiyu is a category rather than one plugin name.

What “raising a fish” means

Chinese community stories repeatedly frame the character as eating tokens, waiting for the user, doing work, reacting to competing models, initiating contact, and asking for rewards after tasks. The expected product is not a fixed corner animation. It is a persistent digital character with an ongoing relationship.

That expectation produces four capability layers:

  • presence: breathing, blinking, walking, sleeping, touching, and dragging;
  • work awareness: thinking, Tool Calls, approval waits, success, and failure;
  • direct interaction: bubble input, voice, task dispatch, and opening a Session;
  • continuity: names, affinity, feeding, memory, diaries, and portable saves.

Current product lines

Product lineExampleMain capability
DSH-native status companionQCYTSN/dsh-dafeiyuNative transparent window and real Session state
Bubble task dispatchyouxin484/dsh-dafeiyu-enhancedSend tasks from the pet to an active Session
Voiced Web characterdsh-pet-dafeiyuChinese voice, subtitles, music, dragging, and linked scenes
3D whaleelyoncatnecoe/whale-pet3D state machine, skins, feeding, music, and dispatch
Persistent aquariumdsh-fishtankMood, relationships, whispers, diary, and JSON migration
Standalone desktop appds-local-pet and othersLocal desktop experience without requiring DSH

Users should not compare screenshots alone. They need to know whether a project actually connects to DSH, which Profile owns it, whether it makes additional model calls, and where character state is stored.

Native DSH status companions

dsh-dafeiyu starts and stops with DSH. Its Host consumes structured Session events while a native transparent window shows work state over editors, browsers, and file managers. The project explicitly says it does not read the screen; reasoning and progress appear only when upstream events provide them.

That makes status claims auditable. “Verifying” should correspond to a real task or tool event rather than a timed random animation. WebUI and official Desktop use separate Profiles, so a market listing must distinguish web from desktop installation.

From status display to task dispatch

An enhanced fork adds a text field above the bubble and routes submitted work to a running, recently active, or otherwise selected Session. The pet becomes a lightweight Agent entry point, but the feature raises important questions:

  • Which Session receives the task, and is that visible?
  • Are its model, workspace, and permission mode clear?
  • Does new input queue, interrupt, or append while work is running?
  • Is a failed draft retained?
  • Can the pet create a new high-privilege Session?

Task dispatch should not automatically grant full access. The pet selects and sends; DSH Session policy must still govern file, command, and network permissions.

Pure companionship, voice, and scenes

dsh-pet-dafeiyu emphasizes character performance: walking, dragging, sleeping, Chinese voice and subtitles, plus a linked scene with another character. Such a pet can avoid reading work content and does not need a model call for every action.

Pure companionship is a valid product line. It has predictable resource and privacy boundaries and avoids spending tokens just to remain alive. Listings should distinguish prerecorded voice, rule-based local lines, and live model generation.

Progression and memory

dsh-fishtank stores mood, energy, people, facts, whispers, and diary entries with import and export tools. dsh-pet already implements feeding, affinity, and rewards tied to completed work rounds.

Healthy progression should reward completed work, continued use, and explicit interaction—not unnecessary model rounds. Otherwise feeding becomes a dark pattern for token consumption.

What every listing should disclose

Each Dafeiyu listing should state:

  1. DSH plugin or standalone application;
  2. Web, Desktop, or native system window;
  3. structured events versus screen/UI scraping;
  4. whether it can dispatch work into Sessions;
  5. access to API keys, balance, conversation text, or local files;
  6. additional model calls and token cost;
  7. storage locations for settings and relationships;
  8. separate licenses for code, character art, animation, voice, and music;
  9. platforms, tested DSH versions, uninstall, and rollback.

Continue with the build-your-own Dafeiyu guide, character assets and transparent animation, Session event integration, memory and progression design, and the permission boundary checklist.

References