Show HN: Tilde.run – 具备事务性和版本控制文件系统的代理沙箱

Tilde started from a simple observation: AI agents are only as useful as the environment they can work in.

Giving an agent a prompt is easy. Giving it the right files, credentials, tools, context, and execution environment, safely and repeatably, is much harder.

Over the past year, we learned a lot from teams trying to run agents in production. 

The most important lesson was that the state (read: the filesystem!) matters far more than people expect.

Agents need more than object storage. 

They need fast local access, reproducible state, isolation, versioning, and the ability to compose data, code, configuration, and intermediate outputs without constantly copying everything around. 

They need infrastructure that feels local, but behaves like a governed, production-grade data system.

That realization brought us back to lakeFS.

Rather than continue tilde.run as a separate product, we decided to focus our efforts on lakeFS and fold everything we learned into lakeFS Mount: an efficient, agent-native, composable filesystem for production AI workloads.

Mount is designed for the way agents actually run: many short-lived environments, large data surfaces, repeated reads, changing working state, and a need to move between experimentation and production without losing reproducibility or control.

It combines the ergonomics of a filesystem with the guarantees lakeFS is built for: versioning, branching, commits, reproducibility, and governance over data at scale.

Shutting down tilde.run was not a step away from agent infrastructure. 

It was the opposite. It was a decision to build it where we believe it belongs: directly into the data control plane.

To everyone who used tilde.run, gave feedback, or helped us understand what production agent workflows really require: thank you. 

Those lessons are now shaping the next phase of lakeFS.

If you came here looking for tilde.run, we’d love to hear what you were trying to build. Whether you’re running agents over large datasets, composing production environments for AI workflows, or looking for a better filesystem abstraction for agent infrastructure, reach out to us. lakeFS’ Mount is where we’re putting this work now, and we’re actively working with teams that are pushing agents into production