Personal AI assistant & agent operations

An assistant that tells you what actually needs you.

kyu.co builds an always-on AI assistant that reads across your chats, email, calendar, and reminders, filters out the noise, and sends one short briefing with the things that need a reply or an action. Nothing is sent or changed without your approval.

Request early access

What it does

One briefing, every source

Messaging apps, Gmail, calendar, and reminders are merged into a morning and evening briefing, plus hourly alerts only when something is urgent.

Only what needs action

Tracks what you already answered, drops stale and automated items, and turns the rest into concrete next steps with dates.

Signal from group chats

Summarizes busy community chats into the few links, tips, and decisions worth your time, with sources attached.

Drafts, not auto-sends

Prepares replies, transcribes voice notes, and researches follow-ups. Sending, booking, and paying always wait for a human yes.

Who it is for

Busy professionals and founders juggling several messaging apps, inboxes, and calendars across countries and languages.
Families who share calendars and need to know whose plan is whose.
Power users running AI agents who want scheduled jobs, model routing, and audit trails without babysitting them.

How we use Claude

Reasoning core: Claude is the main model for reading mixed-language messages, deciding what needs action, and writing the briefing.
Agent operations: Claude drives tool use, including local data collectors, mail and calendar APIs, and a headless browser, inside scheduled jobs with explicit approval gates.
Fit for each task: large models handle judgment-heavy summaries; faster models handle classification and urgency checks, keeping cost per briefing low.

From the blog

All posts →

Notes on local AI, agents, and benchmarks. Latest: Running Qwen3.6 27B dense locally on M4 Max.