Coding agents need somewhere safe and fast to actually run code, mutate filesystems, hit databases, and try risky changes — but existing sandbox options either boot too slowly to iterate, leak state between attempts, or cannot snapshot a full running environment including memory and DB state. Indexable is building purpose-built sandbox infrastructure for AI agents where forking a live environment (files, processes, memory, and databases) happens in milliseconds, so an agent can branch, experiment, roll back, and re-branch as part of a normal reasoning loop.
Under the hood, Indexable runs a custom VMM built on top of KVM that can fork a running sandbox in milliseconds, a content-addressable storage engine with tiered caching and automatic deduplication across tenants, and a storage fabric over libfabric that lets the system move from TCP to RDMA without rewriting application code. That combination unlocks the headline behavior: agents fork real environments, run code, use the terminal, and roll back or branch again in milliseconds instead of spinning up a fresh container for each attempt.
Indexable is part of YC's Spring 2026 (P26) batch, founded by Andrew Gazelka, a former xAI engineer with a high-performance systems and Rust background. The company is based in San Francisco with a team of three, hiring founding engineers, and primary YC partner is Pete Koomen. The bet — that agent reasoning loops will be bottlenecked by sandbox state management rather than model quality — is one several frontier labs are also leaning into.