News for hackers. Stuff that matters.

ajaysheoran2323

AI Sandboxes with Memory

Code execution sandboxes were designed for CI pipelines, not agents. Agents need to remember what they scraped last Tuesday, fork into 5 parallel browsers, and pick up where they left off. That's a different product category. Try sandflare.io

( Read More... | no comments | on HN )
Re:How to build an Agent on Tuesday April 14, 2026 @12:49AM (#47761369)
Building an agent usually means solving three things:

1. Planning 2. Tools 3. Execution environment

Execution is often overlooked. Agents need somewhere to run code, install deps, write files, and continue workflows safely.

We ran into this while building agents, so we built Sandflare — sandboxes with memory that launch in milliseconds.

Agents can spin up environments, run code, and continue later without losing context.

https://sandflare.io

Thanks! The snapshot approach is exactly what we're using — UFFD (userfaultfd) to restore from a pre-booted snapshot, with on-demand page faulting so the VM responds before all memory is loaded. That's how we get to ~300ms.

Sandflare – I built a sandbox that launches AI agent VMs in ~300ms

I've been building Sandflare for the past few months — it launches Firecracker microVMs for AI agents in ~300ms cold start. The idea came from running LLM-generated code in production. Docker felt too risky (shared kernel), full VMs too slow (5–10s). Firecracker hits the middle: real VM isolation, fast boot.I also added managed Postgres because almost every agent I built needed persistent...

( Read More... | 5 comments | 2 points | on HN )