iRaven

IRAVEN // ALL QUIET

iRaven

Building for safety

iRaven is an open research lab for the safe use of AI. It builds its products on the latest academic research and on the results of its own research and experiments, and it publishes them free of charge.

Read the papersOpen the consoleAsk for a tenant

What iRaven is

iRaven is a lab of senior engineers and researchers from around the world, who have come together as volunteers to make AI safer. The lab works on the safety of AI agents and AI assistants: the memory they keep, the models they call, the commands they run, and the computer they work on. In each of these fields it does research, publishes what it finds, and builds a product from it.

How the parts fit together

Apps and their users

  • Your app on a phone or a computer, holding no secret
  • Your server, holding its secret key
  • Your users, each on their own devices

iRaven

  • Muninn
  • Huginn

Tools that serve them

  • Sign-in and devices
  • Usage and ceilings
  • Sealed storage

Model providers

  • Providers that iRaven sets up
  • Your own providers (coming soon)
A user signs in to your app. The app asks iRaven for a model by a name that you chose. iRaven checks the call against the user's ceilings (the most that the user may use in a period), then sends it to a provider with a key that never leaves iRaven. iRaven also keeps the user's memory for your app. To use Muninn and Huginn, ask for a tenant: a place on iRaven for your app and your team, which your team manages in the console.

Papers

The lab publishes its papers here.

The first papers of the lab are being prepared for publication.

All papers