Skip to content
Our productIn development

Credence AI

Run big AI models on the PC you already have.

Large AI models usually need a graphics card that costs more than the rest of the computer. Credence AI is a desktop app that runs open models locally on a normal laptop or PC. It checks your hardware, tells you which models will actually run, downloads them and lets you chat with them, all on your own machine.

01 /How it works

Three steps from install to your own AI.

  1. 01

    Scan your PC

    Credence AI reads your processor, memory, graphics card and free disk space when it opens.

  2. 02

    Pick a model

    Every model gets a label: Recommended, Will run slowly or Won't run. The ones that fit are listed first.

  3. 03

    Chat, offline

    Download it in one click and chat with it on your own computer. No account and no cloud.

02 /Features

Built for people, not just developers.

  • Private by default

    Your chats are saved on your computer only. A Hugging Face token is optional and stays in your system keychain.

  • Streams from your SSD

    Built on the open source Colibri engine, which reads the model from a fast SSD instead of keeping it all in memory.

  • Honest fit score

    No guessing from a spec sheet. You see whether a model will run well on your machine before you download it.

  • Downloads that resume

    Pause, close the app or lose Wi-Fi. Downloads pick up where they stopped and are checked before use.

  • An assistant, not just a chat

    It can search the web, save files and remember what matters to you, and it asks before using a tool.

  • Updates itself

    When a new version is out, the app tells you and installs it in a click.

03 /Roadmap

Where it stands.

  1. Done:

    Idea

    Big AI models need expensive GPUs, so most people never get to run one.
  2. Done:

    Desktop app

    Hardware scan, model downloads and local chat built for Windows and Linux.
  3. Not yet:

    Public beta

    Next
    Open source on GitHub, with free installers anyone can try.
  4. Not yet:

    Launch

    macOS support, more models and a polished first release.
Early access

Try it first.

Get the beta installer before the public launch and tell us what to build next.

Get early access