Will Wright’s Gallium Studios is reviving its long-stalled simulation project, MyMems, by integrating generative AI middleware to transform user-submitted memories into dynamic game environments. The project, originally announced in 2018 as Proxi, faced a funding crisis and mass layoffs in late 2024. Now, the studio is leveraging a partnership with Artificial Agency, a firm specializing in agentic systems, to overcome technical and financial hurdles without becoming an AI research entity itself.

The Architecture of Memory Mining

The core technical innovation in MyMems relies on Artificial Agency’s 'MemDNA' tool, which constructs a personal behavior model and knowledge graph from player inputs. Unlike traditional data mining that tracks clicks or watch history, this system claims to learn from memories locally and only with explicit user permission. Wright praises the 'entirely open-ended' AI systems used to extract narrative elements from typed stories, allowing the game to generate simulated locales and characters that reflect individual player histories.

Infrastructure and Cost Efficiency

For developers watching the bottom line, Wright’s decision to use 'off-the-shelf' AI solutions rather than training frontier models from scratch is a significant infrastructure shift. He explicitly noted that frontier models are 'really expensive' and often overkill for specific game design needs. By utilizing Artificial Agency’s middleware, Gallium avoids the high compute costs associated with custom model training while still achieving the generative depth required for a simulation based on unstructured personal data.

Key Takeaways

  • Gallium Studios revived MyMems after running out of funding in late 2024.
  • The project uses Artificial Agency’s MemDNA to create local knowledge graphs from user memories.
  • Wright emphasizes cost efficiency by avoiding expensive frontier model training in favor of middleware.
  • The system generates simulated characters and locales based on 'entirely open-ended' AI extraction.

The Bottom Line

This is a pragmatic win for indie-scale simulation: using specialized middleware to bypass the prohibitive costs of frontier AI training while maintaining data privacy claims. It proves you don’t need a massive GPU cluster to build complex, personalized generative experiences.