cumulo-nimbus
Nimbus · personal AI, on your desk
A little cloud on your desk, with a mind of its own.
Nimbus is a desk-sized personal AI assistant. Its memory lives on an SD card inside the device, not in anyone's cloud. Bring your own keys for OpenAI, Anthropic or Mistral, or point it at a model you serve yourself. It works fully without us; the cloud is optional, and that's the point.
Small hand-built batches. No spam, only batch announcements.
Two modes, one device
Pick its job: status light, or assistant.
One mode at a time: pick at setup, switch whenever.
An assistant that actually remembers.
Message it on Telegram or just talk to it. Notes, people and projects live in layered memory on its SD card, searchable months back. Give it a hard question and it runs background research on its own, then reports back. It speaks to the big AI labs with your keys, or to a model you serve yourself.
Stop babysitting your coding agents.
You kick off three agents, then lose an hour tabbing between terminals to see if any of them finished. Nimbus consolidates every session, across all your harnesses, into ring light and sound: a segment per agent, readable from across the room, a chime when one is done or stuck waiting on you.
Under the hood
The good parts, shown honestly.
No staged demos. This is how a Nimbus actually behaves.
It runs the whole errand across providers.
One ask becomes several agents on different providers and connectors: the calendar gets checked, the web searched, a shortlist written to Notion, an invite email drafted. You get one message when it is all done, and nothing sends without your OK.
At 03:30, it dreams.
Once a night the day gets reviewed: the keepers move into long-term memory, links form between related notes, the clutter gets compacted. You wake up to an assistant that remembers yesterday properly.
A memory with layers, like yours.
Standing instructions, current work, day-by-day history, and meaning-indexed recall that reaches back months. Every layer lives on the SD card: pull it out and you are holding your assistant's whole mind.
It writes its own abilities. You hold the pen.
Ask for a routine and Nimbus drafts itself a new skill: a readable file you can inspect. It stays inert until you approve it, and you can revoke it any time.
Small rituals, kept faithfully.
A brief every weekday morning, a PR digest on Fridays, the week ahead on Sunday night. Routines fire on the device's own clock, and a drafted routine never runs until you approve it.
Your tools, through your provider.
Gmail, Calendar, Drive, GitHub, Notion, Slack and Linear connect as provider-side tools: Nimbus attaches the connector to the request, and your AI provider executes it. Nimbus itself never holds those logins.
The hardware
Small box, honest spec.
Hand-built in small batches on the Solide S3 board. Every Nimbus ships with all of this.
- Brain
- ESP32-S3, 16 MB flash, 8 MB PSRAM
- Ring
- 45 addressable LEDs
- Display
- 2.9" e-paper with a click knob, or 2.8" touch screen
- Voice
- Microphone and speaker
- Memory
- SD card. Your data stays on it.
- Power
- Battery, three power modes
- Radios
- Wi-Fi, plus encrypted BLE for the notifier link
- Models
- OpenAI, Anthropic, Mistral, or any OpenAI-compatible endpoint you host
Bring your own brain.
Your own keys to the labs, or any OpenAI-compatible endpoint you serve yourself: Ollama, vLLM, llama.cpp. The device does not care whose cloud it is, or whether there is one.
Sovereign by default
You choose exactly how much cloud you want.
Every rung is opt-in, and we tell you plainly what we can see on each one. The device never needs us to work.
Planned pricing
Simple tiers, no surprises.
We are pre-launch: these are the planned prices, and the waitlist hears first when each one goes live.
The device
$129 DIY kit, batch 1
- Hand-built, numbered batch
- Works fully without our cloud
- Free software updates
Nimbus Cloud
$65/year
- Your device's screen, from anywhere
- No port forwarding, no setup
- The device stays the server
Hosted Nimbus
$19/month
- A Nimbus of your own, no hardware
- Metered AI allowance included
- Export your memory any time
Small batches. Honest cloud. Your AI.
Join the waitlist, and tell us what you'd want a Nimbus for. It genuinely shapes what gets built first.
You're on the list. We'll write when the next batch opens.