AgentENV

AgentENV runs isolated environments for AI agents across your own servers. It's for teams building parallel agent workflows or reinforcement learning systems that need many separate sandboxes.

• Isolated environments: Firecracker microVMs run agent workloads across machines, with OCI-compatible images loaded on demand rather than copied to every host in advance.
• Snapshots and forks: Capture memory and filesystem changes incrementally, then fork a running environment into independent sandboxes that explore different paths from the same state.
• Pause and resume: Idle environments release CPU and memory, while snapshot-backed environments can boot or resume in under 50 milliseconds.
• Storage beyond local disks: A bounded local cache holds frequently used image data, while snapshots persist in S3-compatible object storage or a shared distributed filesystem.
• E2B compatibility: The HTTP API works with the standard E2B Python and TypeScript SDKs, so existing integrations can target a self-hosted server.

AgentENV is open source under MIT and supports Docker Compose and Kubernetes deployment. Firecracker execution requires Linux kernel 6.8 or newer with KVM access. Memory ballooning returns reclaimable guest memory to the host so more environments can share a server.

The API authenticates requests but doesn't encrypt traffic itself. It needs a trusted network or HTTPS provided by a reverse proxy or load balancer.