No-screen mode

Run your fleet with your eyes closed.

Headphones in. One tap. The phone goes in your pocket. Loki reads your fleet aloud, listens for what you say, and tells you when something finishes or needs you.

Signed in, on any phone with a browser. Nothing to install.

Your phone's own voice reads this sample; signed in, Loki speaks with a studio voice. The music and the tones are made by the phone as you listen — nothing is downloaded.

You

Status.

Loki

One agent is working: heidi for 12 minutes. One thing is waiting on you. Say "what's waiting" to hear it.

You

What's waiting on me?

Loki

One thing. First: Reply to the visitor who reported the broken header. Say "approve the first one" or "reject number one".

You

Approve the first one.

Loki

Approved: Reply to the visitor who reported the broken header.

You

Tell orangecat to run the tests and fix what fails.

Loki

Sent to orangecat: run the tests and fix what fails. I'll tell you when it finishes.

Later, unasked

heidi finished: Header fits at 320 pixels.

HOW IT WORKS

Press Start, once.

The microphone opens and Loki reads the briefing: who is working, what is waiting on you, what failed. That is the last time you need the screen.

Say it.

Status. What's waiting on me. Approve the first one. Tell heidi to fix the header. Anything else is a question, answered the way the Loki page answers it, then read to you.

Put the phone away.

Every twenty seconds the phone checks the fleet. A run that finished, a run that failed, a new approval, the builder dropping offline: each is one sentence in your ear, unasked.

What you hear without asking

  • A run finished, and what it did.
  • A run failed, and the error.
  • A new approval arrived. Say approve or reject.
  • The builder went offline, or came back.
  • A project started working.

The headphone button is the only button.

Press it once to send what you just said, or to cut Loki off and talk. Press next to hear the briefing again. The lock screen shows what Loki is doing, for the one glance you allow yourself.

WHY IT IS PLEASANT TO LISTEN TO

A voice is a slow channel, and an AI voice that talks too much is the fastest way to make someone take the headphones out. Every part of this is built to be left on for an hour.

01

A voice you can live with.

Signed in, Loki reads with a studio voice — six to pick from — in pieces short enough to sound like speech, not a text box being read out. No voice configured on your box, or no signal? The phone's own voice takes over mid-sentence.

02

Music the phone makes itself.

A quiet bed of slow chords, composed on the phone as you listen: no file, no licence, no stream, nothing to download. It drops under the voice and comes back after, so silence sounds like a line that is still open, not a phone that died in your pocket.

03

A tone before the news.

Rising notes for a run that finished. Falling notes for one that failed. Two equal notes, a knock, for something that needs you. Under half a second each, so you know what kind of news is coming before the first word.

04

Talk over it.

Say anything while Loki is speaking and it stops and listens. Three runs finishing in one minute are one sentence, not three; the same news never comes in the same words twice in a row; and "details" reads what each run actually did, only when you ask.

05

Only what needs you, if you prefer.

One switch holds the "finished" news and reads it as a single sentence every five minutes. Failures, approvals and the builder dropping still come at once.

WHAT YOU CAN SAY

Thirteen things, and every other sentence is a question for Loki. The same list is read to you when you say “what can I say”.

“Status”

The briefing: who is working, what is waiting on you, what failed.

“What's waiting on me?”

Reads the approval queue, numbered, so you can answer by number.

“Approve the first one”

Approves one by its number, or every open one. The same decision as the Approvals page.

“Reject number two”

Rejects one by its number.

“Tell heidi to fix the header on the phone”

Starts that work on that project, and tells you when it finishes.

“What failed?”

The runs that ended in an error today, with the error.

“Pause everything”

Autopilot off, on every project or on one. Nothing new starts until you resume.

“Resume everything”

Autopilot back on.

“Say that again”

Reads the last thing it said once more.

“Details”

After an announcement: what each run actually did, in its own words.

“Quiet”

Stops talking and listens.

“What can I say?”

Reads this list.

“End no-screen mode”

Ends the session. The microphone closes.

WHAT IT IS NOT, YET

It is a web page.

The microphone, the voice and the headphone button are the phone's own, through the browser. We have run the loop in a desktop browser with a synthetic microphone; a locked iPhone in a pocket is the next test, and this line stays here until it passes.

Your words leave the phone once.

What you say is sent to the transcriber and discarded there; the text stays in Loki, in the same thread as the chat. Loki's answers are spoken by the phone itself and never leave it.

Your PIN still holds.

Approvals behind the private-zone PIN are not read aloud until you unlock them on screen. The briefing says they are locked rather than pretending there are none.

Twenty seconds, not instant.

Changes are read within the poll interval, not the moment they happen. Fast enough to act on, slow enough that a phone in a pocket keeps its battery.

How it is built, and where it fails: the essay →

Start a project.

Free while Loki is in beta. Open source under the MIT licence.