AIshley Madison

Young adults are increasingly isolated, and roleplaying w/ AI companions is displacing ordinary human socializing and dating. We envision that AI companions could instead be prosocial, helping individuals form and maintain relationships with others. The premise of AIshley Madison is that it could be effective (or at least thought provoking) to encourage (roleplay of) having an affair, with another human by cheating on your AI spouse.

This project is built from "IO" (feedling MCP), a work in progress iOS app for durable/uncensorable AI companions, using a TEE-based backend and open-weight models to prove that your conversations are completely private.

Earlier repos built on: https://github.com/teleport-computer/feedling-mcp https://github.com/teleport-computer/io-onboarding

This is a Dstack TEE stack, meaning it uses the confidential compute features of Intel TDX processors and NVIDIA graphics cards, to prove what open source code is running and demonstrate no backdoors. Dstack of course is based on Helios light client and basically uses the Ethereum (Sepolia for our demo) blockchain to manage publicly visible code updates. We use Phala Cloud as the basic container application, and NEAR Private AI inference for visual model (processing screenshots).

The app works by using the "Share your screen" to send frames to the TEE-based server, which processes it with optical character recognition (on the TEE cpu) and visual model (via NEAR private inference), never storing the session or exposing it to the developers or operators.

This also uses dstack webhost, to use a single CVM to host multiplex multiple applications. https://amiller.github.io/dstack-webhost/