Skip to content

Press kit

Everything you need to write about Mellix.

For journalists, partners and analysts: our story, the logo, approved wording, the fonts and the colours. Please use them as they are.

Story

Mellix, in brief.

Mellix builds small decision models for the routine decisions inside an organisation's services, starting with the Myles family. Given a situation, a question and the possible answers, a Mellix model returns a probability for every option in a single pass. The models run on the servers, laptops and GPUs an organisation already operates, keep only the decisions they can make within an error rate certified on held-out data, and hand the rest to a person. Mellix is a UK company, and its models are trained and evaluated on UK-owned hardware.

Company
Mellix Limited, registered in England and Wales, company number 17551068
Incorporated
29 October 2026
Registered office
1 Parkside Court, Greenhough Road, Lichfield WS13 7FE
Models
Six sizes: Myles-22M, Myles-150M and Myles-400M run on a CPU; Myles-G Edge, Myles-G 12B and Myles-G 31B run on one GPU
Ways to get it
The hosted Decision API, the CPU models by request in the console, or a self-hosted licence per deployment
Website
mellix.com

Words

Lines you can lift as they are.

  • The name

    Mellix

  • Tagline

    The decision layer for the enterprise

  • In one line

    Small, calibrated decision models that run on the CPUs you already own, inside your network, with an error rate you sign off before go-live.

  • In two sentences

    Mellix builds small decision models, starting with the Myles family, that give every option a probability in one pass and run on hardware an organisation already owns. Each deployment keeps only the decisions it can make within a certified error rate and hands the rest to a person.

  • The headline number

    Myles-400M is 91.8% correct on 1,400 held-out decisions on our frozen evaluation set, measured in September 2026.

  • About Mellix

    Mellix builds small decision models for the routine decisions inside an organisation's services, starting with the Myles family. Given a situation, a question and the possible answers, a Mellix model returns a probability for every option in a single pass. The models run on the servers, laptops and GPUs an organisation already operates, keep only the decisions they can make within an error rate certified on held-out data, and hand the rest to a person. Mellix is a UK company, and its models are trained and evaluated on UK-owned hardware.

Mellix
One word, with a capital M. The company is Mellix Limited.
Myles
The model family, not the company. Write the models as Myles-22M, Myles-150M, Myles-400M and Myles-G.
What it is
Decision models, not chatbots. Built in the UK.

Fonts

Three free fonts, each with one job.

  • Newsreader

    Headlines. Regular 400.

    The decision layer for the enterprise

    Newsreader on Google Fonts
  • Geist

    Everything you read, the menu, the buttons and the logo word. Regular 400, Medium 500 for the logo.

    Every option gets a probability, in one pass.

    Geist on Google Fonts
  • Geist Mono

    Labels, figures and code. Regular 400, usually small.

    91.8% · 27 ms · Myles-400M

    Geist Mono on Google Fonts

Colours

Deep blues on warm paper.

  • Logo blue

    #111B2B

    The logo, headlines and the main button

  • Navy

    #0E1B2C

    Dark bands, image cards and the footer

  • Indigo

    #172B76

    Accents: label markers, the active step and indigo image cards

  • Grey

    #5E5C57

    Supporting text

  • Stone

    #EBE9E4

    Light panels

  • Paper

    #F5F5F1

    The page

Calls to action

Two buttons, one for each reader.

Talk to us
For enterprise, self-hosting and security questions. Goes to the contact form on the Enterprise page.
Get an API key
For developers who want to call the models through the Decision API. Goes to sign-up in the console.

Press contact

Talk to the team.

For interviews, quotes and anything not here, email press@mellix.com or use the contact form and choose Press and media. We reply within 24 hours on working days.