What is saobase?

SAObase is an experimental database designed for AI agents rather than for humans typing SQL by hand. The central idea is that every write to the database goes through a single controlled gate before it becomes official truth. Agents do their exploratory work on separate branches, similar to how a programmer might work on a branch in version control, and only when that work is merged back into the main record does the system check policy rules, verify the change, record who made it, settle any budget the agent was allowed to spend, and permanently log the entry, all as one atomic step called the Gate. The project grew out of a workshop on agent-oriented data systems and an earlier project called WormBase, which introduced the idea of a single, permanent, tamper evident ledger as the only source of truth, with every other view of the data generated from that ledger. The README also compares the approach to Supabase, the popular database and backend platform, arguing that just as Supabase built authentication and other services directly into the database rather than bolting them on separately, SAObase wants to build governance and safety controls directly into the database for agent systems, plus an outer layer that manages policy and a client toolkit that helps agents track budgets and decisions. The project is currently an early preview release, not a finished product, with an internal checklist of required features being tracked and verified before anything is considered complete. To try it, a developer needs the Rust programming language installed. After cloning the repository, they can install a command line tool called sao and run a built in demo, or manually create a store, branch off proposed changes, and merge them through the Gate using a series of commands. The system does not support rewriting history, only adding corrective entries on top of it. The codebase is organized into several Rust crates that separate the core storage engine, input and output handling, a small SQL like language, the command line interface, testing tools, and a formal model used to check correctness.