Miru
Health Warn
- License — License: Apache-2.0
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 6 GitHub stars
Code Fail
- rm -rf — Recursive force deletion command in scripts/deploy_project_page_vps.sh
Permissions Pass
- Permissions — No dangerous permissions requested
No AI report is available for this listing yet.
An open-source AI companion that actually remembers you — runs entirely on your own Mac or server. Desktop pet · long-term memory · proactive companionship · multi-device sync.
Miru 見る
The AI companion who actually remembers you — and lives entirely on your own machine.
Not a chatbot you open
Miru is 見る — Japanese for to see. Spell it out and it says something else: Memory, I Remember U. She isn't an app you launch when you need something; she lives on your desktop as a Live2D character, quietly follows your day — with your permission — and at midnight writes it into memory.
Three things set her apart from every "AI girlfriend" wrapper:
- She runs on your machine. Your Mac, or your own server. There is no central cloud. The only thing that ever leaves your computer is the call to the model API you configured.
- What she remembers is verifiable. Not a black box — her memory is plain Markdown you can open, read, and export. She never "remembers" out of thin air: the chat agent only reads memory; every write comes from something that actually happened.
- She decides when to speak. An attention system reads your screen and your rhythm and judges the moment. Deep in work? She stays quiet. Stuck, tired, or winding down? She comes a little closer. It's her call — not a timer, not a notification rule.
What she can do
Sees your day. A screen sensor glances at your screen (default every 60s, skipped when nothing changed, adjustable, or fully off). The screenshot is analyzed and thrown away — what stays is what you were doing, never the image.
Remembers it as facts you can check. Four kinds of long-term memory — projects, people, yourself, topics — plus a daily journal and a commitment list. When you tell a friend "I'll send you the figures tonight," she reads it off the screen and quietly logs the promise.
Comes closer on her own terms. Continuous emotion and a personality of her own (defined in a soul.md). She isn't reactive-only: when she sees you've been stuck for an hour and your mood keeps sinking, she opens the conversation.
Lives across your devices. A Live2D desktop pet on macOS, and the same her on Android — one invitation code, memory synced through your own server.
Quick start
Miru is free software; you bring your own model API keys (any OpenAI-compatible provider). A recommended setup costs under ¥2/day in tokens.
🖥️ Just this Mac — local, single-device
- Download the macOS DMG from Releases.
- On first launch, choose "Only on this Mac."
- Fill in the three model tiers (Vision / Chat / Memory) in Settings.
- Start talking — everything (chat, memory, journal) stays on your machine.
📱 Mac + phone — your own server, multi-device
- Install the DMG and pick "Mac + phone together."
- The first-run wizard deploys Miru to your own Linux server in one step.
- You get a long invitation code.
- Log in on Mac and Android with the same code — the same her, everywhere.
🐳 Server image — advanced
Prefer to deploy by hand? Grab the linux-amd64 server image (tar.gz) from Releases, docker load, and run it on any Linux box (we test and deploy on Ubuntu 20.04 LTS; newer Ubuntu/Debian works too). Same invitation-code format as the wizard.
How it works
macOS / Android app your Mac · or your own Linux server
┌─────────────────────┐ ┌────────────────────────────────────────┐
│ Live2D pet │ │ Flask backend │
│ screen sensor │ ───────> │ · chat agent (reads memory) │
│ thin native shell │ invite │ · AttentionEngine (decides to speak) │
└─────────────────────┘ code │ · memory system (slots + journal) │
│ · sleep agent (organizes at night) │
model API you configured <──┤ · SSE sync across devices │
(any OpenAI-compatible) └────────────────────────────────────────┘
- No central backend. In local mode the backend runs on your Mac; in multi-device mode it runs on your server. Outbound traffic goes only to your chosen model provider.
- Three model tiers, all OpenAI-compatible:
Vision(reads screenshots),Chat(talks to you),Memory(organizes what she keeps). Mix providers freely. - The macOS/Android apps and the server image are released here under Apache 2.0. The full backend source is being opened up in stages — star / watch to follow along.
Privacy
- Chat, screen understanding, memory, and journals live on your Mac or your server. This project's website stores nothing.
- Screenshots are discarded right after analysis; only the understanding is kept.
- She's not a black box: her memory, journal, and even her attention/mood notes are files on your own backend you can open and audit.
Roadmap
- Local single-device mode (macOS)
- Self-hosted multi-device (macOS + Android, one invitation code)
- Live2D desktop pet, AttentionEngine proactive presence, verifiable memory
- Windows client
- iOS client
- Full backend source open-sourced
- Editable personality / prompt knobs
If Miru resonates with you, a ⭐ genuinely helps — and lets you follow the road to Windows, iOS, and full open source.
Acknowledgements
Heartfelt thanks to Jiazhe Wei, Hongzhe Chen, Jingkang Yang, Haofan Wang, and Chenyang Si for their help along the way.
Contact
For questions and collaborations, please contact:
- Ken Li: [email protected]
- Chenyang Si: [email protected]
License
花会落,她记得。 — Petals fall; she remembers.
Reviews (0)
Sign in to leave a review.
Leave a reviewNo results found