Your whole life, in a closed loop — Arandu ingests every source, connects it by what matters, and hands you only what's relevant, when it's relevant. Entirely on your Mac.
Calendar is light today — room to clear what's open. Sandra wants help with gov.br and hinted at a visit. Amor's birthday is Wednesday, in 3 days.
Syncing data · 1m 54s ago
↑ vs week (avg 4)
Sandra Mara · 1d open
4 on track · 3 attention
Arandu runs entirely on your Mac. Local LLM inference is the default and nothing leaves the machine. If you want frontier-grade reasoning, you can turn on Arandu Cloud — but only after a consent screen that spells out exactly what gets sent, and every call it makes is written to a tamper-evident log you own.
Calendar, mail, contacts, notes, files. Everything stays on disk in a local database on your Mac. We never touch your data over the network.
Local inference is the default: nothing leaves your Mac at all. Turning on Arandu Cloud takes an explicit, one-time consent screen that says plainly what goes out — your conversation, including the personal context Arandu has learned. No dark pattern, no buried toggle.
Each cloud request is appended to a SHA-256-chained audit log on your machine — which model, when, how many tokens. The chain is tamper-evident: break a link and verification fails. Read it and verify it yourself, any time, in the app.
Not a chatbot. Not a productivity dashboard. A continuously running personal AI that learns your life and quietly keeps it organized.
A morning summary in your language. People mentioned, money discussed, dates coming up, threads needing a reply.
Threads, tasks, and habits collapsed into one ranked stream sorted by what matters most today.
Pin a horizon ("schedule volleyball reservations"), pin a why ("regular sports + social"). Arandu links every task to a goal.
Write your own — schedules, prompts, delivery to WhatsApp / email / push. Or use the 33 system agents that ship with the app.
Calendar, Mail, Contacts, Files, WhatsApp, Notion, Linear, Jira, Strava, Oura — all indexed locally on first sync.
Every firewall decision and every cloud call — cryptographically chained. Verify the chain anytime in Settings.
Privacy isn't a checkbox on a page. It's a chain of design decisions you can inspect — in source code we publish under Apache 2.0.
Pay only for what you use. Local LLM inference is free. Cloud LLM inference is billed per token — $10 minimum top-up. Earn $5 in credits for every friend you refer.
The packaged Arandu app. Runs LLM inference locally by default, with the option to route to Arandu Cloud when you want more reasoning power. New features land here first.
$5.00 / 1M tokens out. One rate, every model — we pick the right model for each job, you never have to. No subscription. Minimum top-up $10. Set a monthly cap.
What does $10 buy? Roughly 5 million tokens — around 200 daily briefs, or ~60 deep multi-step reasoning answers. Refer a friend to earn $5 in credits each. Set a hard monthly cap so you never overspend.
When someone signs up for Arandu Cloud with your link and makes their first $10 top-up, you both get $5 in credits — roughly 2.5 million tokens each. No cap on referrals.
github.com/vdittgen/arandu. Build it yourself and you get a fully local assistant with no cloud path in the codebase. What's closed is the Pro overlay: the cloud gateway, the account and billing service, and the client code that talks to them. We'd rather say that plainly than claim "fully open source" and hope nobody checks.