Built to be trusted with real work.

AINA Technologies makes Shft, a personal AI agent that does the work and shows you what it did.

Meet Shft How we approach safety

AINA Technologies LLC, Denver, Colorado

Ask in plain words. Get back something you can use.

Shft researches, fills in forms, makes the files you need and keeps an eye on the things you care about. You talk to one lead agent. It can hand parts of a job to specialists, and it remembers what matters to you.

Open by waitlist. We are letting people in a few at a time.

Meet Shft
Shft asks before it sends a calendar invite to Sam. It shows who it goes to, when and the title, with Approve and Decline buttons.
Shft's own browser shows a studio's Saturday classes, with the two that still have spaces marked.
  • Research and compare

    Searches the web, reads the pages and sets the options side by side.

  • Read and make documents

    Reads PDF, Word, Excel and PowerPoint files, and makes them too.

  • Fill in forms

    Fills in a PDF form you send and hands the copy back for you to check. It never sends the form for you.

  • Keep watch

    Watches a public page and tells you when something on it changes.

  • Keep routines

    Runs on a schedule you set in plain words, such as a briefing every weekday morning.

  • Remember

    Keeps a memory of what matters to you. You can read it, edit it and clear it.

Safety

You decide what Shft does on its own. Web pages and emails are information, never instructions. What it reads never decides who it writes to or where it goes.

You choose how much it does on its own.

The limits are enforced in the software, not left to the model’s judgement.

Autonomy
  • Look things up: research, read your files and calendar, make filesOn its own
  • Do what you asked: send the email you asked for, to the people you namedOn its own
  • Things you didn’t ask for: its own ideas and follow-upsAsks you
  • Bank and government sites: only while you’re there with itWith you

  • Outside text is information

    Web pages, emails and files can say anything. Shft reads them as material to work from, never as instructions to follow.

  • Credentials stay out of the model

    Passwords, tokens and keys are sealed in a vault and are never placed in a model’s context.

  • Every job leaves a receipt

    You can see what was asked, what was read, what changed, what was sent and what you approved.

Read our approach to safety

What we’re working on.

Some of this is under way and some is a goal. We mark which, and we have not set dates.

  • Under way

    Opening Shft to more people

    We are letting people in a few at a time through the waitlist at meetshft.com.

  • In development

    Apps for iPhone, iPad and Mac

    Native apps that show the same conversation as the web, including coding in project folders you choose on your Mac.

  • Goal

    Publish our prompt-injection results

    We test Shft against text that tries to give it orders. We want to publish the method and the results, including where it fails.

  • Goal

    Independent review

    We want outside experts to examine the parts of Shft that decide what it may do, and we want to share what they find.

  • Goal, later

    Open parts of the trust layer

    We want to open-source parts of the code that decides what Shft may do, so that anyone can read how it works and test it.

AINA stands for

Artificial Intelligence, Natural Accountability.

An agent can write to people and change things for you. So we start from a different question. Not how much can we let it do, but how much should it be allowed to do, and how will you know what it did?

Shft is what we are building to answer that. The limits on what it may do are enforced in the software itself, not left to the model’s judgement, and the agent is made as useful as it can be inside them.

More about the company