cd /news/large-language-models/desk-pet-a-local-first-desktop-pet-p… · home topics large-language-models article
[ARTICLE · art-50168] src=github.com ↗ pub= topic=large-language-models verified=true sentiment=↑ positive

Desk-Pet – a local-first desktop pet powered by MiniCPM5

OpenBMB released Desk-Pet, a local-first desktop pet application powered by the MiniCPM5 language model, allowing users to chat with a tiny companion on their desktop without requiring remote inference after initial setup. The app features agent-aware reactions that detect coding activity from tools like Cursor and Claude Code, providing task narration and idle alerts. Available for macOS and Windows, it includes a guided first-launch setup for environment checks and model downloads.

read3 min views1 publishedJul 7, 2026
Desk-Pet – a local-first desktop pet powered by MiniCPM5
Image: source

English | 简体中文

A local-first desktop pet powered by MiniCPM. Download, complete the guided setup, and chat with a tiny companion that lives on your desktop.

Local by default— after the model is downloaded, everyday chat runs on your machine.** Zero manual setup**— first launch guides you through environment check, model download, and warm-up.** Desktop companion**— open a floating chat bubble, talk with MiniCPM, and keep the pet on screen while you work.** Agent-aware reactions**— the pet can react to coding activity from tools such as Cursor, Claude Code, and Codex.** Task narration**— when a coding-agent session finishes, the pet summarizes what the AI just did in a speech bubble, so you can catch up at a glance.Idle alerts— if a coding agent has been waiting for your input, the pet plays a bell animation and sound to get your attention.** Auto-detect agents**— the app scans your machine for installed coding agents and prompts you to connect them in one click.** Smart model download**— the app can download from Hugging Face or ModelScope and choose the better source for your network.** Persona support**— switch or import character adapters from** Settings -> MiniCPM**.

Item Recommended
macOS 14.0+, Apple Silicon (M1/M2/M3/M4), about 2 GB disk space
Windows x64 with Vulkan support, about 2 GB disk space
Network Required on first launch unless you already have a local model file

macOS Apple Silicon is the primary tested platform. A Windows installer is also available — feedback is welcome.

macOS

  • Go to Releasesand download the latestMiniCPM Desk Pet-*-arm64.dmg

. - Open the DMG and drag MiniCPM Desk Pet intoApplications

. - Launch the app and follow the setup guide.

If macOS blocks the first launch, right-click the app and choose Open. If needed, remove the quarantine flag:

xattr -cr /Applications/MiniCPM\ Desk\ Pet.app

Windows

  • Go to Releasesand download the latest.exe

installer. - Run the installer and complete the wizard.

  • Launch the app and follow the setup guide.

MiniCPM Desk Pet includes a complete first-launch guide:

Environment Check -> Model Download -> Model Warm-up -> Ready to Use

The default model is MiniCPM5-1B-GGUF. You can let the app download it automatically, or choose an existing local .gguf

file.

Use the floating chat bubble to talk with MiniCPM from your desktop. Once setup is complete, your normal conversations do not need a remote inference service.

Useful shortcuts (macOS uses Cmd

, Windows uses Ctrl

):

Cmd/Ctrl+Shift+M

— open or close the MiniCPM chat bubbleCmd/Ctrl+Shift+T

— show or hide thinking modeEsc

— close the bubble when input is focused

MiniCPM Desk Pet can stay beside your workspace and react to coding-agent activity: thinking, working, finishing tasks, waiting for attention, or going idle.

The MiniCPM settings page lets you:

  • download the default model or choose a local model file
  • rerun onboarding
  • manage character/persona adapters
  • restart the local model runtime when needed

The app includes a neko-style persona adapter. You can switch adapters or import your own from Settings -> MiniCPM.

  • Broader Linux validation.

  • More persona presets.

  • Clearer model download diagnostics and retry guidance.

  • Faster first launch and smaller app footprint.

  • Richer desktop-pet narration for long-running coding sessions.

  • The primary tested release target is macOS Apple Silicon. Windows is supported with a bundled installer; report issues if something does not work on your setup.

  • First launch requires an internet connection unless you provide a local model file.

  • Response speed depends on your chip, memory pressure, and selected model.

  • Coding-agent reactions depend on each tool's integration behavior and may vary by version.

For development setup, packaging, and repository layout, see docs/development.md.

  • Desktop pet UI is based on rullerzhou-afk/clawd-on-desk. Full attribution is listed in.NOTICE.md

  • Model weights come from the OpenBMB MiniCPM model family and are downloaded separately.

  • The bundled neko persona uses the neko30k dataset (liumindmind/NekoQA-30K) for fine-tuning data.

This repository is distributed under GNU AGPL-3.0-only.

MiniCPM model weights are downloaded separately and governed by the OpenBMB MiniCPM Model License. Artwork, third-party code, and datasets keep their own notices; see NOTICE.md and

.

clawd-on-desk/NOTICE.md

── more in #large-language-models 4 stories · sorted by recency
── more on @openbmb 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/desk-pet-a-local-fir…] indexed:0 read:3min 2026-07-07 ·