AI agents, each with its own computer

Your AI team
works
the
night shift.

Build agents for coding, research, customer support, voice calls and narrated decks. Each one works on its own computer. On a big job, it keeps working until a separate check confirms the goal is met.

Watch one night

All on one subscription

Scout Driftline launch

When should we launch our rain shell in Seattle, Portland, Vancouver, San Francisco and Denver? Use the last ten years of rain data, and have a plan ready by morning.

working06:31 PM to 07:12 PM
Working…Browsing open-meteo.com…Running python fetch.py --years 2016-2025Running python rain.py weather-2016-2025.csvBrowsing localhost…Writing brief.md
Read the Open-Meteo archive docs1 action2 actions
Browsing open-meteo.comrunning
open-meteo.com/en/docs/historical-weather-api
Working in the browser…
Browser · runningOpen
Pull ten years of daily weather for five cities1 action3 actions
Running python fetch.py --years 2016-2025running
$ python fetch.py --years 2016-2025
Running…
Terminal · runningOpen
Count the rainy days in each month1 action2 actions3 actions
Running python rain.py weather-2016-2025.csvrunning
$ python rain.py weather-2016-2025.csv
Running…
Terminal · runningOpen
Browsing localhostrunning
localhost:8000/heatmap.html
Working in the browser…
Browser · runningOpen
Write the launch brief1 action4 actions
Writing brief.mdrunning
Writing brief.md
Files · runningOpen

Worked 41m 06s·12 actions·1 180.00 credits

Launch Vancouver in September, Seattle and Portland in October, San Francisco in November and Denver in April.

Send a comment to the agent...Assign a task or ask anything...
Deep

Enter queues · ⌘Enter steers now

Presentation designer Northwake Driftline Rain Shell Launch Deck

Make a 6-slide launch deck for Northwake's new Driftline rain shell. Cover the problem (most rain shells wet out within an hour), the jacket (3-layer, 310 g, fully taped seams, $189), who it's for (day hikers and bike commuters), the launch plan (pre-orders open October 7, stores October 21), and close with one line to remember.

working
Working…Searching the web for “how fast 2.5-layer rain shells wet out”Building the presentation…Editing slide…Working…
Look up what the slides will cite1 action3 actions
Searching the web for “how fast 2.5-layer rain shells wet out”running
Searching the web…
Search · running
Build the six slides1 action11 actions
Building the presentationrunning
Building the presentation…
Presentation · running
Review each slide and fix what it finds2 actions4 actions6 actions8 actions10 actions12 actions26 actions
Reading presentation
Editing sliderunning
Slide 1 · Tightened the tagline
Presentation · running
Reading presentation
Editing sliderunning
Slide 2 · Made the 60-minute figure the focus
Presentation · running
Reading presentation
Editing sliderunning
Slide 3 · Evened the four feature columns
Presentation · running
Reading presentation
Editing sliderunning
Slide 4 · Matched the two audience cards
Presentation · running
Reading presentation
Editing sliderunning
Slide 5 · Aligned the launch dates
Presentation · running
Reading presentation
Editing sliderunning
Slide 6 · Kept the closing line to one sentence
Presentation · running
Narrate all slides in presentation1 action27 actions
Actionrunning
Working…
Presentation · running

Worked 10m 44s·67 actions·923.96 credits

Here is the 6-slide launch presentation designed for Northwake’s new Driftline rain shell. Narrated all 6 slides with tailored scripts matching each slide's core message.

Send a comment to the agent...Assign a task or ask anything...
Auto

Enter queues · ⌘Enter steers now

Scout Driftline stockists

Find outdoor stores in each of our five launch cities that could stock the Driftline shell, and send me a shortlist.

On it. I’ll give each city to a helper and tell you as they finish.

Shortlist stores in five cities

6m 12s

Step 2 of 3 · helpers working21 stores in five cities, ranked by fit · 1 artifact

Hide the stepsStop
Working with 5 helpers · 2 at a time · 0 doneWorking with 5 helpers · 2 at a time · 1 doneRan 5 helpers · 186 creditsStop all

Running 5 helpers · 2 at a time — roughly 5× the credits of a single run (est. 380 credits)

Seattle storesReading store pagesDeep · 14 credits
Portland storesSearching the webDeep · 9 credits
Portland storesFound 5 storesDeep · 36 credits
Vancouver storesQueued · waiting for a helper slot
Vancouver storesReading store pagesDeep · 3 credits
San Francisco storesQueued · waiting for a helper slot
Denver storesQueued · waiting for a helper slot

For “Shortlist stores in five cities” it just split the work across 5 agents running in parallel.

One of the parallel agents finished — Portland stores.

All five are in. The shortlist is on your screen: 21 stores, ranked by how well they fit.

Scout · Live4:12 1 task running1 task done SpeakingEnd
Type into the call…
Deep
Nightwatch Checkout is failing

Something’s broken in checkout. Find it and fix it before morning.

working06:42 PM to 08:31 PM
Running npm test -- cartEditing src/cart.tsEditing src/cart.test.tsRunning npm test -- cart && gh pr create --fill
Reproduce the failure in the cart tests1 action8 actions
Running npm test -- cartrunning
$ npm test -- cart
Running…
Terminal · runningOpen
Fix the coupon total in src/cart.ts1 action2 actions9 actions
Editing src/cart.tsrunning
Editing src/cart.ts
Files · runningOpen
Editing src/cart.test.tsrunning
Editing src/cart.test.ts
Files · runningOpen
Run the tests and open a pull request1 action6 actions
Running npm test -- cart && gh pr create --fillrunning
$ npm test -- cart && gh pr create --fill
Running…
Terminal · runningOpen

Worked 1h 49m·23 actions·640.00 credits

PR #214 is up with three new tests. CI is running.

Changed 2 files+31−1

Send a comment to the agent...Assign a task or ask anything...
Deep

Enter queues · ⌘Enter steers now

Ledger Daily payments

Loop

working07:00 AM to 07:01 AM
Working…Using Stripe…Running PythonBuilding table “Payments, 24 Sep”Using Slack…
Pull yesterday’s payments from Stripe1 action
Using StripeExternalrunning
Using Stripe…
Stripe · running
Total them by status1 action
Running Pythonrunning
summarize("payments-2026-09-24.json", by="status")
Running…
Python · runningOpen
Post the summary in #finance1 action2 actions
Building table “Payments, 24 Sep”running
Building table “Payments, 24 Sep”…
Table · running
Using SlackExternalrunning
Using Slack…
Slack · running

Worked 1m 12s·4 actions·38.00 credits

Posted in #finance: 214 payments yesterday, $18,420 in total. The post lists the 3 that failed.

Send a comment to the agent...Assign a task or ask anything...
Auto

Enter queues · ⌘Enter steers now

Launch timing from ten years of rainDocument Open

In private beta. Built by a small team, and you can talk to us directly.

One agent, one night

Here’s one night.

Follow Nightwatch from a voice note at 18:42 to a verified fix before midnight, then the rest of the team through to morning.

18:42 pass 1

A voice note on your way out.

“Something’s broken in checkout. Find it and fix it before morning.”

The agent turns it into a goal it can check, and a plan it works through in passes.

19:06 pass 1

It reproduces the bug like a customer would.

In its own Chromium it adds the sweater to the cart, applies the coupon and presses Pay. The payment fails. Now it has proof.

You can take over the browser mid-run. Control goes back to the agent on a timer, or when you hand it back.

19:48 pass 1

It reads the logs in its own terminal.

The stack trace points at the coupon total, and the last change to that file added percentage coupons. That’s the bug.

20:31 pass 1 ends

It opens the pull request. Then a separate check reviews the pass.

PR #214 is up. When the pass ends, a separate check compares it with the goal. CI hasn’t finished and nothing is re-tested yet, so it books another pass.

22:26 pass 2 ends

A separate check decides when it’s done.

In pass 2 the agent waited for CI, then ran checkout with a percentage coupon, a fixed coupon and none. This time every check passes, so the goal is met.

22:40 you call in

You call it on the drive home.

Hand it the next job out loud: write the release note for PR #214 and post it in #releases. Ask how far it’s got and it tells you. Hang up, and it keeps working.

Nightwatch Checkout is failing

Something’s broken in checkout. Find it and fix it before morning.

working06:42 PM to 08:31 PM
Working…Working in the browser…Working…
Reproducing the bug like a customerReproduce the bug like a customer1 action6 actions
Working in the browserrunning
everline.shop/checkout
Working in the browser…
Browser · runningOpen
Finding the cause in the logsFind the cause in the logs3 actions
Fix the coupon total and open a pull request
Check checkout with each coupon type

Worked 1h 49m·23 actions·640.00 credits

PR #214 is up with three new tests. CI is running.

Changed 2 files+31−1

Deep · Goal pass 2 of ≤5

08:33 PM to 10:26 PM

Worked 1h 53m·14 actions·450.00 credits

Goal met. Checkout works with all three coupon types, and PR #214 is ready for you to merge.

Write the release note for PR #214 and post it in #releases.

On it. I’ll draft it from the pull request and post it there.

Write the release note for PR #214

48s

Step 2 of 3 · drafting the note

Show the stepsStop
How far along is it?

It has read the pull request and is drafting the note now.

Checkout completes with percentage coupons, fixed coupons and no couponNext pass at 08:33 PM

Pass 2 of ≤5640/1500 credits

Stop

Verifier: The fix is in PR #214, but CI hasn’t passed and checkout hasn’t been re-tested with each coupon type.

Plan 0/41/4 Preparing computer… Goal done
Nightwatch · Live1:08 1 task running SpeakingEnd
Send a comment to the agent...Assign a task or ask anything...Type into the call…
Deep

Enter queues · ⌘Enter steers now

Recreated from recorded runs. Timings compressed.

Front desk

It answers the phone, too.

Give an agent a phone number and a designed or cloned voice. It answers from your Knowledge, replies by voice when a customer sends a voice note, and keeps every call’s transcript in its thread.

WhatsApp, Telegram, Slack, email, SMS, website chat and phone calls, with more channels on the way.

Calls and SMS run on your own Twilio number.

Front desk Duplicate charge refund

Customer · Phone

Hi. I was charged twice for order 9140. Can I get the $640 back?

Looked up order 9140 in Shopify

Found two $640 charges in Stripe, two minutes apart

You’re right: order 9140 was charged twice at 01:12. I’ve flagged the second charge for a refund, and you’ll get an email once it’s done. I’m texting you the case number now.

Texted case 9140-R to the caller

The whole night

It wasn’t the only one on shift.

Every agent has its own machine, so they work side by side without stepping on each other. A Loop wakes them on an event or a schedule.

An example night. Nightwatch’s prompt is the voice note above; the other prompts are examples.

Your machine

Or put it on your own machine.

By default an agent’s computer runs in our cloud. Install our runner on a Linux machine you own, and the agent works there instead: in your repo, with the tools already installed. The runner only connects out, so there is nothing to open on your network.

Your repo and your tools

It works in the folder you choose, with what is already installed there. What it writes stays on your disk for its next run.

Sites that turn away cloud servers

On Business and up, an agent’s web reads can go out through your own connection. The run says which way each read went.

Nothing to open

The runner connects out on port 443. In Confined mode, the default, the agent can’t see your SSH keys or cloud credentials, and can’t reach your local network.

Pro includes one machine and Business five. Ubuntu or Debian, amd64 or arm64.

How it works

The Auteryn workspace, recreated. The machine, the repo and the timings are made up.

Rules

It stops where you tell it to.

Autonomy is a dial on each agent. Every plan step and every action streams into the run timeline, and a Loop that fires at 04:30 follows the same rules as a run you start by hand.

Ledger Double charges 1

Loop

working
Using Stripe…
Used Stripe2 actions
Using StripeExternalrunning

Order 9140 was charged twice, $640 each. I’m refunding the second charge.

WriteExternal71h 49m left

Stripe create refund

{
  "payment_intent": "pi_3QxT…9140",
  "amount": 64000,
  "reason": "duplicate"
}

Asking because autonomy is set to Supervised.

Real output

What the starter agents made.

The windows above are recreations. These are recorded runs: we made each agent from its template, typed one prompt into its chat and kept what came back. Nothing was edited, only converted for the web. Open one to see any follow-up it needed and the original files.

Recorded on a paid plan. AI-generated. Northwake is a made-up company.

See them all on the agents page

Integrations

Works with your stack.

20+ native integrations and channels, plus a featured catalog. Each agent carries only the Integrations its job needs, and runs on Auto or, on Pro and up, on the model you pick.

Native integrations, channels and sandbox CLIs

The MCP catalog, vendor-hosted and self-hosted

More native integrations are on the way. If yours isn’t here yet, point the agent at any MCP-compatible service you run.

Pricing

Start free. Go Pro for $19.

Every plan gives each agent its own computer. Pro adds voice calls, phone calls and video. One subscription covers the whole team.

Launch pricing is on. Your first 3 months of Pro or Business cost less on monthly billing.

Free

$0

Try it on real work.

Start free

No credit card needed

What it can do

  • Work on its own computer, with a browser
  • Research, documents and decks
  • Code and pull requests
  • Loops
  • Connect your apps
  • Answers from your docs
  • Start from a ready-made agent
  • 1,000credits a month
  • 1 hcomputer time a month
  • 1agent

Voice calls start on Pro.

Pro Recommended

Launch price

Regular price $29Launch price $19/ month

For your first 3 months, then $29. Save $30.

Keeps at one task for days or weeks.

Start with Pro

Cancel anytime

Everything in Free, plus

  • Voice calls with your agent
  • Answers calls on your own Twilio number
  • Clone up to 3 voices for it to speak in
  • Narrated decks, shared as video
  • Video, voiceovers and podcasts
  • Skills from screen recordings, with no monthly limit
  • Answers on Slack, WhatsApp, Telegram, email and your website
  • Your own Linux machine as its computer
  • 15,000credits a month
  • 25 hcomputer time a month
  • 3team members

Enough for up to 8 hours of voice calls a month.

Business

Launch price

Regular price $99Launch price $79/ month

For your first 3 months, then $99. Save $60.

Everything in Pro, for your team.

Start with Business

Cancel anytime

Everything in Pro, plus

  • Your whole team, up to 10 people
  • Up to 5 of your own machines
  • Reads sites that block cloud servers, through your own machine
  • Clone up to 10 voices
  • Priority support, 24 h response
  • 60,000credits a month
  • 100 hcomputer time a month
  • 10team members

Enough for up to 33 hours of voice calls a month.

Enterprise

Custom

Custom contracts for isolation, SSO and support.

Talk to us

Everything in Business, plus

  • Custom credits and limits
  • Audit logs
  • Unlimited cloned voices
  • On contract: SSO / SAML
  • On contract: 24/7 support, dedicated CSM
  • The launch price is applied at checkout, no code to type
  • Top-up packs on Pro and Business never expire

Compare every plan

1 credit is $0.001. All prices in USD. The launch price applies to monthly billing and to a team’s first subscription. Every model is served by a US-based provider. Open models are served from US servers, never by the lab that made them.

Give it tonight’s problem.

Or talk to the founder, or email founder@auteryn.ai.