The Runtime is a self-hosted agent engine built on a single Postgres instance and licensed under Apache-2.0. It treats every autonomous agent like an employee: no job runs without intent, authority, evidence, and settlement — and every job must prove all four when it's done.
Autonomy without accountability is a liability. The Foxbra Runtime exists to make agent autonomy safe to deploy: it wraps every action in a governance contract so that power is always bounded and outcomes are always provable.
It runs on infrastructure you already understand. One Postgres engine holds state, evidence, and the ledger of settlements — no sprawling service mesh, no opaque vendor cloud. Clone it, run it, audit it.
Each of these is required before a job starts and verified after it ends. Together they turn “the agent said it did it” into “here is the proof.”
The outcome the job must produce, declared up front and carried through every step so work can't silently drift from the goal.
Explicit, bounded permission — exactly which resources an agent may touch, and nothing beyond it. Power is granted, never assumed.
Content-addressed proof of what actually happened, so any claim about the work can be independently checked after the fact.
Signed receipts on completion: obligations met, within limits, on the record. The job closes with a verifiable outcome.
State, evidence, and settlement live in one database you control. Operationally boring on purpose.
Permissively licensed. Read the source, fork it, ship it in your own product with no lock-in.
The ledger is the product. Every action is reconstructable, so compliance isn't bolted on afterward.
Pre-1.0 and open for you to run today. Bring your own models via the GRIP protocol and drive it from Studio or your own client.