SQLite for Rivet Actors

Nathan Flurry and Nicholas Kissel are building something that makes database architecture look like it actually should in 2025. Rivet Actors is an open-source framework for stateful workloads—think of it as Cloudflare Durable Objects but open-source and you actually own your data. They just dropped SQLite storage for actors, and the pitch is simple: every single actor gets its own database. Not a table. Not a row. A whole SQLite file.

This matters because the current state of AI agents and multi-tenant apps is a disaster. Most teams are cramming agent memory, conversation history, embeddings, and user state into a shared database and praying their RLS policies hold up. Rivet says "screw that"—give each agent, tenant, user, or document its own isolated database and scale horizontally. We're talking millions of independent databases running at the edge. For AI agents, that means per-agent memory that doesn't bleed into other sessions. For SaaS, real isolation without the RLS hacks that every developer regrets at 2am. For collaborative docs, each document is its own little database with built-in multiplayer support.

Nathan previously built Krunker and Addicting Games, so he knows a thing or two about scaling real-time stuff for tons of users. Nicholas is right there with him. The whole thing is Apache 2.0 licensed and running on GitHub. It's the anti-vendor-lock infrastructure play—Cloudflare Durable Objects does similar things but you're stuck in their ecosystem. Rivet gives you the same architecture without the leash.

They're based wherever open-source lives these days (remote-first, probably), and they need contributors, users, and people building wild stuff with it.