Skip to content

The action layer

From a sentence to a finished job — with your permission at every step.

Most AI assistants can answer questions. Fewer can actually do things. And of those that can, most operate through rigid, pre-defined integrations that require explicit commands to trigger.

hu is designed around a different paradigm: an action layer that can translate natural conversation into real outcomes, while maintaining the user control and transparency that trust requires.

From conversation to execution

You might say, “Move dinner with Maya to next week.” That isn’t a command — it’s intent. The action layer turns it into steps: work out what you mean, check what it’s allowed to do, draft the change, and then either ask you or — if you’ve allowed it — go ahead and tell you what it did.

From intent to action

  1. Intent“Move dinner with Maya to next week.”
  2. UnderstandWork out what you mean.
  3. Permission checkCheck what it’s allowed to do.
  4. DraftPrepare the change.
  5. Ask or actAsk you — or, if you’ve allowed it, go ahead and tell you what it did.
Every action passes a permission check before anything happens.

The key is that you don’t have to think about the mechanics. You express what you want to happen; hu figures out how to make it happen.

Permission architecture

The action layer operates within a careful permission system. Some actions — like setting a reminder for yourself — are generally safe to execute immediately. Others — like sending a message to someone — require explicit confirmation. Still others — like making a purchase or sharing sensitive information — will require additional authentication. That tier is next.

You’ll be able to tune these permissions over time: let hu handle routine calendar changes without asking, or require your yes before anything goes out.

hu tasks waiting for a yes: “Email Maya about Sunday dinner” needs you before anything is sent.hu tasks waiting for a yes: “Email Maya about Sunday dinner” needs you before anything is sent.Product UI · sample data
In the app todayHand it a job. It asks before it acts.hu plans the steps and waits for your yes before anything is sent, booked or paid. Every action is logged.Built

Acting on your behalf

One of the most powerful — and most sensitive — things we’re building is hu acting on your behalf in communication with others. Once mail is connected, hu will draft, show you the draft, and wait. Anything that goes out after your yes will be logged, so you can review it later.

This capability requires deep trust, so we’re building it with safeguards: clear disclosure that responses come from hu, faithful representation of your preferences and tone, and detailed logging so you can review every action taken.

The integration challenge

For hu to take actions across your digital life, it needs to connect with the tools you use — calendar, email, messaging, task management and more. Each integration brings its own technical challenges and permission considerations.

We’re building these integrations carefully, with user control at every step. You choose what to connect. You can see what hu can access. You can revoke permissions at any time. The goal is a system that can genuinely help manage your life — but only to the extent you want it to.

Today the action layer works with web search and the mail, messengers and calendars you connect — so hu can run your everyday channels itself.

All writing

See it in the app.

Your own hu, ready in a minute.