Matilda runs on hardware
we own in Melbourne

Chat, write code or hand off a whole task. Matilda asks before it does anything with side effects.

Chat and Code share one plan, from $25 a month.

One plan for Chat and Code

Starter

Free

Daily chat and a couple of coding tasks a week, plus a 5M token welcome grant for Code

Daily chat, plus a couple of coding tasks a week

A couple of coding tasks a week, plus a one-time 5M token welcome grant

2M tokens a week

Standard

$25a month

Around 9 coding tasks or 2,800 chat messages a week

Around 2,800 chat messages a week

Around 9 coding tasks a week, about 2 a workday

8M tokens a week

Super

$125a month

Around 54 coding tasks or 17,000 chat messages a week

Around 17,000 chat messages a week

Around 54 coding tasks a week, about 10 a workday

48M tokens a week, then pay-as-you-go if you choose

Chat and Code draw on one weekly pool. Prices in Australian dollars, including GST. For a team, see Organisations.

Questions about Matilda

Is Matilda free?

Yes, to start. Starter is free and covers Chat and Code, with a one-time 5M token welcome grant for Code. Standard is $25 a month and Super $125, and the pricing page compares them. The API is priced per token, and you can get a key yourself.

Where does my data go?

Matilda processes and stores your conversations on MC-2, the cluster we own and run in Melbourne, and an organisation can have its own instance there or in its own data centre. The trust and security page lists each commitment and the policy it’s written in.

Where can I use it?

Matilda Chat runs on the web at matilda.com, on iPhone and Android, and in the desktop app for macOS, Windows and Linux. Matilda Code runs in your terminal, the desktop app and CI.

What can’t it do yet?

It can be too cautious, it can miss context, and parts of it aren’t yet as fast as we want. Tell us when it gets something wrong, because that’s the feedback we want most.

Can my organisation run its own Matilda?

Yes. We start from the open-weights model that suits your work, train it on your material and align it to your requirements, and you run it in your own instance and hold the weights. See Enterprise.