Sandboxes

Ephemeral compute,
one API call away.

Run any container image + command once, get the exit code and output. Perfect for AI agents running untrusted code, one-off jobs, and build steps. The flat-rate alternative to e2b, Modal and Vercel Sandboxes.

Agents need to run code — safely, and without you babysitting a server

When an AI agent needs to execute untrusted code, transform a file, or run a quick task, spinning up and tearing down infrastructure is friction. koigrid Sandboxes run a container image and command once, isolated, with memory and time limits, and return the output — all from a single API call the agent makes with its own token. No server to keep alive, no cleanup.

How koigrid compares

koigridDIY VM
PricingFlat ratePer VM-hour + idle
Run once, auto-teardownYesYou manage it
Memory / time limitsYesManual
Isolated per runYesDepends
One API callYesNo
Agent-native (token + scopes)NativeNo

From public docs, 2026. koigrid runs the image on the same runner fleet as Apps and Jobs, isolated, and records each run’s output.

Compute that appears and disappears

One-shot runs

docker run --rm for you: image + command in, exit code + logs out.

Bounded

Memory and timeout limits per run — nothing runs away.

Agent-native

An agent runs a sandbox with its token and scopes — perfect for untrusted code.

Same fleet

Runs on the runners that power your Apps and Jobs — no separate service.

Auditable

Every run is recorded with its output, status and duration.

Flat rate

One price — no per-second billing to reason about.