Tarit - Fastest hypervisor for self-hosted AI agents and RL
Show HN: Tarit – a hypervisor which is 2x faster than firecracker
Tarit is a high-performance microVM platform designed specifically for secure, ephemeral sandboxes in AI agent and Reinforcement Learning workloads. Built on Rust-vmm, it boots real hardware-virtualized VMs in milliseconds, offering kernel-level isolation that surpasses shared-kernel containers. The system includes a standalone VMM and a multi-node orchestrator, taritd, enabling rapid task execution, live snapshots, and efficient resource management. With sub-second startup times and features like warm pools and per-key usage metering, Tarit provides a robust, scalable solution for developers needing secure, bursty compute environments without the overhead of traditional virtualization.
Tarit boots a real hardware-virtualized VM in milliseconds, runs a task inside it, and tears it down, giving each sandbox kernel-level isolation instead of a shared-kernel container boundary.
- philips
Where does the headline come from? The README metrics say there are no published benchmarks.
- miduil
Not "yet" with "virtio-balloon", so basically where firecracker was for quite some time. With that you can only grow, but not shrink memory if I understand correctly. So useful for "I'll run and terminate" but not for "I'll run next to others with varying runtime and memory usage".
Will be interesting to see how https://github.com/rust-vmm/rust-vmm is going to develop, sounds like something like that could have a lot of use-cases for privilege separation but also embedded testing frameworks.
- DenisM
Not clear where 2x claim in the headline came from, I don’t see it in the linked site.