Run the same work on fewer servers.

Tau makes the libraries your services spend their CPU in faster on your machines, and proves that each one behaves exactly like the original.

17% less CPU
storage_server
read_block
snappy::RawUncompresssnappy
checksum
copy
network
other

Illustrative profile. Snappy decompression runs 1.59x faster than Google Snappy compiled at its fastest for that CPU (measured on Intel Sapphire Rapids, Silesia corpus). Its output is proved to be exactly the decompression the Snappy format defines, and tested identical to Google Snappy's on 4,034,025 streams with no differences.

Your services

Where the CPU time goes

A one-minute profile, or a connected repository

Your machines

The CPU you run on

e.g. GCE C3 or AWS m7i: Sapphire Rapids, AVX-512

Your libraries

The exact versions you ship

e.g. Pillow 13, libjpeg-turbo, zlib

Tau

Optimise the libraries your CPU time is in, for your machines

Every version is proved before you get it

You get

A faster library

Same interface. A drop-in replacement

You get

A proof

For every input, the same output as the original

Your services

Where the CPU time goes

A one-minute profile, or a connected repository

Your machines

The CPU you run on

e.g. GCE C3 or AWS m7i: Sapphire Rapids, AVX-512

Your libraries

The exact versions you ship

e.g. Pillow 13, libjpeg-turbo, zlib

Tau

Optimise the libraries your CPU time is in, for your machines

Every version is proved before you get it

You get

A faster library

Same interface. A drop-in replacement

You get

A proof

For every input, the same output as the original

Drop-in

No code changes and no migration. You swap in the faster library.

Lower compute costs

The same work on less CPU, on every machine that runs it.

Same behaviour

Proved identical to yours for every input, so there is nothing new to trust.

One plan.

$200 a month, per company

Up to 3 projects. Cancel any time: the faster libraries and support end with the subscription.

Get started
  • Up to 3 projects, with unlimited people
  • Faster versions of the libraries your CPU time is in, each proved identical to yours
  • Kept up to date as your code changes
  • Yours to use while you subscribe, with support

See where your CPU time goes.

One command, or connect a repository. Nothing is sent until you say so.

Get started