Foxbra is building the accountability layer for autonomous AI. We think the next decade of software will be run by agents — and that the only agents worth trusting are the ones that operate like accountable employees: bounded, obligated, and provable.
The industry raced to make AI capable. Far less effort went into making it answerable. Give a capable agent broad access and you get speed — and a black box you can't audit when something goes wrong.
We started Foxbra to close that gap. Not with a policy PDF or a compliance checkbox, but with an engine: a runtime where every job is a contract, every action is bounded, and every outcome carries proof.
Agents should be treated like employees, not magic. An employee has a job to do (intent), a scope they're allowed to work within (authority), a record of what they did (evidence), and a sign-off when it's finished (settlement). Software agents deserve nothing less — and neither do the people who deploy them.
The Runtime is Apache-2.0 and GRIP is an open spec. Accountability works best as a shared standard, not a moat.
We'd rather show a receipt than ask for trust. Evidence is the product, not the marketing.
One Postgres engine, permissive licensing, no lock-in. Powerful ideas should run on calm foundations.
Deploying a workforce, self-hosting the Runtime, adopting GRIP, or curious about Fennec? We'd like to hear what you're building.