June 2025 Summaries

Large Language Models (LLMs) can be integrated with Postgres and several tools to create a personalized AI assistant capable of long-term memory retention and autonomous actions. This system utilizes an LLM-owned database schema, scheduled tasks, real-time web searches, and integrations with external tools like Zapier MCP for extended functionalities, such as managing emails, calendars, and smart home devices. It maintains three types of memory for conversation continuity—Message History, Semantic Memory using vector search, and Structured Memory with SQL data. The assistant operates through scheduled prompts powered by pg_cron and can interact with users via a Telegram Bot or other interfaces. Use cases include tracking portfolios, daily runs, meals, customer feedback, and article bookmarking, with the flexibility to adapt behavior based on user preferences. The project encourages users to build their own AI assistants while maintaining control over their data, with implementation guidance available through a GitHub repository.

Today, Sugu, the co-creator of Vitess, is joining the Supabase team to build Multigres: Vitess for Postgres. Vitess is a database clustering system that was originally built at YouTube and adds features such as sharding, connection pooling, query routing, resiliency, and failover to scale MySQL databases. Multigres is a new proxy designed specifically for Postgres, with the goal of providing compatibility and scalability while prioritizing tradeoffs in a way that aligns with the Postgres community's values. The project aims to provide developers with a gradual on-ramp from simple connection pooling to sharded solutions as they grow into petabyte-scale databases. Sugu believes that Vitess provides the most promising foundation for addressing Postgres' scaling limits and is excited to join Supabase to make this happen, with no fundamental changes planned to the way Supabase operates today.