$200 free credits · no card required
Firecracker microVMs your agents can provision, use, and tear down via API. Per-second billing. E2B-compatible SDKs. MCP-native.
Your agent needs to run code, install packages, process files. Today, that means fragile Docker setups, shared machines, or expensive cloud VMs that take minutes to provision.
Everything an AI agent needs from compute, nothing it doesn't.
Firecracker microVMs boot in milliseconds. Warm pools keep environments ready. Your agent never waits.
Each sandbox runs in its own microVM with dedicated kernel. Not containers — real VM-level isolation with cryptographic erasure on teardown.
REST API + TypeScript SDK + Python SDK. MCP tool discovery so Claude, Cursor, and other agents find you automatically.
Pay only for what you use. Real-time metering. $200 free credits to start. No overprovisioning, no surprises.
Your agent provisions a sandbox, runs code, gets results, tears down. Four API calls.
POST /v1/sandboxes — a Firecracker microVM boots with your chosen image and resource limits.
Run code, install packages, read/write files. Full Linux environment with network access.
Stream stdout/stderr in real-time. Download artifacts. Query the filesystem.
DELETE /v1/sandboxes/:id — VM destroyed, storage cryptographically erased, billing stops.
The same technology AWS Lambda uses for isolation, purpose-built for agent workloads.
Start free. Scale with per-second billing. No contracts.
$200 in credits. No card required.
For production agent workloads.
Dedicated infrastructure, SLAs, BYOC.
$200 free credits. No card required. First sandbox in under a minute.