Hey Product Hunt 👋 Oz here, co-founder of Treeverse. We've been building lakeFS for years, and today we're launching tilde.run, a second product from the same team, aimed at a problem we kept seeing land in our laps. When we built lakeFS, the bet was that data deserves the same safety nets code has long taken for granted: branches, commits, atomic merges, the ability to roll back when something goes sideways. We spent years giving teams those primitives for data lakes, and then agents showed up. It became clear the people handing Claude the keys to production had jumped thirty years backward in operational maturity. An agent rewriting a customer table is a DROP TABLE with extra steps if there's no branch to land it on and no commit to revert. The stories piled up fast: coding agents force-pushing over teammates, data agents "cleaning up" production buckets, research agents quietly burning five-figure API bills. None of it is exotic. It's the predictable result of giving autonomous software write access with no isolation and no rollback. That's a problem we already know how to solve. tilde.run wraps an agent inside the same primitives we built for data, composable filesystems, branches, atomic commits, mediated egress, so its mistakes become survivable. lakeFS isn't going anywhere; tilde.run is what happens when you point the same ideas at a new caller. Would love to hear how you're handling agent blast radius today, and happy to answer anything about the architecture, the lakeFS lineage, or where we're headed!

tilde.run

Serverless sanbox for agents, with a versioned filesystem.

Serverless sanbox for agents, with a versioned filesystem.