Products

Two things we are building for ourselves.

Everything else on this site is work we did for a client. These two are ours. Neither has shipped, and we are not taking signups yet — this page exists so you know what is coming and can tell us if it is useful to you.

The companion

Not a chatbot. Something that finishes the task.

It runs on your machine, keeps a machine of its own, remembers what you told it months ago, and does all of it in the language you actually speak.

Give it work like you would a person.

Ask in a sentence. It takes the task from start to finish, keeps the context of how you work, and comes back when it needs your approval — not before.

It works your computer

Signs in to your tools and uses them the way you would — apps, files, browser, forms. Whole tasks, end to end.

Your computerWorking
your apps
  1. Opened the supplier portal
  2. Pulled the invoices out of your folder
  3. Filled the form the way you would
  4. Filed the batch, left one for you

It has its own computer

Long jobs and anything risky run on a machine of its own — its own desktop, its own browser — so yours stays yours.

Its computerIsolated

nothing touches your desktop

It remembers

A long-term memory vault. What you tell it once, it still knows months later — and you can read, edit, or wipe it.

Memory vaultUpdated

They only sign annual contracts, and Dana approves.

Noted for next time

It speaks your language

Understands and replies across South African languages, out loud, with the channel staying open between turns.

VoiceListening

isiZulu

Products

In development

Both demos below are interactive and entirely local to your browser. Nothing here is connected to a model — they show how the products behave, not what they output.

You said

Sort this month’s invoices and file the VAT return.

  1. Opening Downloads
  2. Found 14 invoices · 3 duplicates
  3. Renaming by supplier + date
  4. Filing into /Accounts/2026-09
  5. Filling VAT201 in the browser
  6. Stopped for your approval
Working on your desktop
In development

A companion that works your computer, not a chatbot that talks about it.

It runs on your machine and can actually operate it — open the apps, move the files, fill the forms, finish the task. It keeps a virtual machine of its own for anything you would rather it did not run on yours. It remembers what you told it last month. And it does all of that in the language you speak, out loud, without you touching the keyboard.

  • Operates your computer directly — apps, files, browser, forms, whole tasks end to end
  • Its own virtual machine, so long jobs and anything risky run isolated from your desktop
  • A long-term memory vault: what you tell it once, it still knows months later
  • Understands and replies across South African languages, not English-only with a translation layer
  • A continuous voice mode — stay in the conversation and talk, no press-and-hold between turns
Optimise for
HostLatencyThroughput$/1MUptime
highveld-gpurouted310ms74 t/s$0.4299.71%
aurora-compute240ms92 t/s$0.5999.94%
karoo-infra180ms118 t/s$0.7899.89%
meridian-aidegraded890ms31 t/s$0.5597.20%
Request
curl https://api.asyncrouter.co.za/v1/chat/completions \
 -H "Authorization: Bearer $ASYNC_API_KEY" \
 -d '{"model":"llama-3.3-70b","route":"price"}'

Illustrative - host names and figures are placeholders

Coming soon
AsyncRouter

One endpoint. Every model, every provider.

A routing layer for language models: one API key and one endpoint in front of many models and the providers that host them. Point at a model, and the router picks a healthy provider to serve it — falling back automatically when one degrades, and showing you what each call cost.

  • One OpenAI-compatible endpoint across model families and hosts
  • Route by price, latency, or availability — and fail over when a provider degrades
  • Per-call cost and latency visible across every model you use

Neither has launched. Tell us what you would use it for and we will come back to you when it is ready.

Want early access?

Tell us what you would use it for. We will come back to you when there is something to try.