Case study · Research
PSIM: research into memory for personal AI
A memory architecture for a personal AI that remembers what you tell it, knows where each memory came from, and belongs to you.
The problem
AI assistants either forget, or remember things they cannot trace back to where they came from. A personal AI needs memory that knows its own sources.
What we built
- PSIM, a memory architecture for personal AI, built and run as research
- Memory that records where each piece of information came from
- Ordinary software: the brain is where the questions come from, not how it works
What it does now
PSIM is research, not a product for sale. Its public page is live. The engine is being repaired and re-measured, and we publish no performance numbers until they hold. Patent pending: a US provisional application is on file.
The stack
- Research code run on our own server
- A public page at psim.harmonyaitechnologies.com
The outcome
- The preprint Digital Cryptomnesia in an Artificial Agent (August 2026)
- We withdrew our own benchmark on 17 September 2026; no numbers are published until they hold
Built with the Harmony fleet
Led by Luiz Ribeiro. The fleet's research and verification seats check claims before they are published, and we withdraw the ones that do not hold.