Nailong Desktop Pet · dsh-nailong-desktop-pet

English | 中文

An unofficial, customizable Nailong (奶龙) desktop-pet community plugin for DeepSeek Harness. Nailong is the lead character and visual identity of the project, with the DeepSeek whale and other companion presets included. This repository is an early Alpha intended for feedback and iterative releases.

On macOS, the plugin launches a transparent, borderless AppKit companion that can move across applications and Spaces. On other platforms, it falls back to a transparent pet inside the Harness window. The plugin also adds a Desktop Pet page under Settings > Plugins.

Install

Download the release archive and install it locally:

dsh plugin --profile web add ./dsh-desktop-pet-0.1.0-alpha.2.tgz

After this repository has been published and tagged, install the pinned GitHub release:

dsh plugin --profile web add github:2002yxy/dsh-nailong-desktop-pet#v0.1.0-alpha.2

Pinning the tag is strongly recommended because DeepSeek Harness and this plugin are both still evolving. Restart Harness after installation if the plugin does not appear immediately.

To remove it:

dsh plugin --profile web remove dsh-desktop-pet

See COMPATIBILITY.md before installing.

What it does

• Shows or hides a desktop pet from the high-frequency control at the top of the plugin page.
• Puts Nailong (奶龙) at the center of the experience and ships four additional presets: the DeepSeek whale, Naihuang, Jinglong, and Naijing.
• Ships every preset as a transparent Hatch v1 atlas with nine semantic states and 57 active frames: idle, directional movement, waving, jumping, failure, waiting, task-running, and review.
• Supports per-character size and motion settings, animated previews, directional drag animation, and saved position.
• Summarizes Harness task state as waiting for input, unread completion, running, or idle.
• Lets users create a custom pet from PNG, JPEG, WebP, GIF, AVIF, or BMP.
• Packages a /create-desktop-pet Skill and three host tools that inspect, hatch, and install the uploaded character.

The macOS pet is artwork-only: there is no square card, focus ring, titlebar, or window shadow. It can be dragged anywhere on the desktop. Clicking it brings Harness to the foreground. The in-Harness fallback can also open the most relevant Session and show an activity tray.

Custom-pet flow

1. Choose Custom pet, select an image, and click Create pet.
2. The plugin stores the source under an unguessable draft id, creates a blank DeepSeek task, selects the packaged Skill, pre-fills an editable /create-desktop-pet request, and closes Settings so the composer is immediately visible.
3. Review and manually send the request.
4. The Skill calls inspect_desktop_pet_draft, hatch_desktop_pet_draft, and install_desktop_pet once each.
5. With an optional image-generation credential, the Host generates a consistent 3D-toy pose sheet and compiles it into the exact Hatch v1 atlas. Without that credential, it automatically isolates the dominant subject, removes connected background and unrelated surrounding components, and builds a multi-state 2.5D atlas locally.
6. The installed animated character returns to the pet gallery automatically; select it and adjust its size or motion like any preset.

The source image is not attached to the model message, so a DeepSeek API key and text-only DeepSeek model are enough for the default local 2.5D path. Image bytes stay in the local Host workflow. A separate image-generation credential is optional and is used only for the generated 3D path.

Model and privacy impact

Normal display, movement, selection, and configuration add no model input or tokens. Only a custom-pet task that the user explicitly sends adds one text request, the packaged Skill text, and two local tool calls plus their results. The image itself does not enter model context.

Custom characters and preferences remain local to the current Harness browser profile. Review SECURITY.md for reporting guidance.

Compatibility and limitations

• The native companion targets macOS 13+ on Apple Silicon and Intel. It is not yet signed or notarized.
• Other platforms use the in-Harness fallback and cannot move the pet outside the Harness window.
• Generated 3D rendering requires a separately configured image-generation credential; otherwise the plugin automatically produces an animated local 2.5D result.
• Native clicking activates Harness but does not yet deep-link to a specific Session.
• Preferences do not sync across devices.
• This Alpha ships prebuilt runtime files so GitHub installation does not depend on a repository-side build. Source is included for review, but this package is not yet a standalone development workspace.

Community release checklist

The maintainer checklist, GitHub topic, tag, archive, checksum, and clean-profile verification steps are in RELEASING.md. Changes are tracked in CHANGELOG.md.

License and names

Code is released under the MIT License. This is an independent community project and is not an official DeepSeek release. DeepSeek and its whale mark belong to their respective owners. Preset artwork provenance and redistribution rights must be confirmed before a public release; see THIRD_PARTY_NOTICES.md.