Introduction¶
The Web UI for DeepSeek Harness (DSH) provides context for Agent runs, but more intuitive feedback is sometimes needed. The DSH Agent Pet plugin places a pet in a corner of the Web UI, and its actions follow changes in Agent state. It is maintained by Noah-wang and addresses the need to confirm the current Agent behavior through visual feedback during long-running sessions or debugging.
Core Features¶
The core capability of the plugin is to generate and display frame-by-frame animated sprite sheets. It maps Agent lifecycle states to concrete actions: breathing while idle, lifting its head while thinking, running while executing tools, tilting its head while waiting for confirmation, jumping upon completion, lowering its head when an error occurs, and so on. The sprite sheet is a 6x8 grid and contains 8 states. The built-in default pet, Momo, is a hand-written SVG image and does not depend on the network.
Installation¶
The plugin requires Node.js >= 22.19.0.
dsh plugin --profile web add dsh-agent-pet
Restart the DSH Web UI after installation.
Usage¶
The plugin provides three ways to have a pet:
- Use the default appearance: After installation, you can use the built-in SVG pet Momo without any configuration.
- Generate one via Studio: In the “Pet” section on the settings page, enter a name, appearance description, and art style (presets include auto, pixel, sticker, etc.), then click “Start Drawing”. This consumes an API Key to generate a new 6x8 sprite sheet.
- Import an external sprite sheet: Export a prompt from another image-generation AI (such as Midjourney), generate the sprite sheet, and upload it. The plugin automatically recognizes transparent spacing in the image, so rows and columns do not need to be specified manually.
During local debugging, you can switch states through the API:
curl -X POST http://127.0.0.1:3080/api/agent-pet \
-H 'content-type: application/json' \
-d '{"state":"running_tool"}'
Configuration¶
The plugin supports configuration through the patch layer of a profile. All options are optional:
apiKeyEnv: credential reference name for the image service (default:OPENAI_API_KEY).model: image generation model (default:gpt-image-2).baseURL: image service endpoint (default:https://api.openai.com/v1).quality: generation quality (low/medium/high).dataDir: directory where generated results are stored.petFile: path to a custom pet definition file.idleAfterMs: number of milliseconds after completion or error before returning to idle.
Privacy and Dependencies¶
The plugin does not read conversation content, tool parameters, tool results, API keys, or other credentials. It only observes Agent lifecycle events and tool names.
In terms of dependencies, the plugin has only a React peer dependency and no other runtime dependencies. It accepts only valid PNG image files and rejects scripts or HTML.
Conclusion¶
The current version is 0.5.0. This pet is suitable for developers who need visual feedback on Agent state in the DSH Web UI. The source code and documentation are available on GitHub.