forkd

forkd is a microVM sandbox runtime for AI agent fan-out that spawns isolated children from a warm parent snapshot using copy-on-write memory.

forkd is a microVM sandbox runtime designed for AI agent fan-out. It works by booting a Firecracker parent VM once, importing the user's runtime and dependencies, and pausing it to disk. Each child is a separate Firecracker process that uses memory mapping to inherit the parent's memory through copy-on-write, avoiding the latency of cold-booting kernels. It provides KVM-level isolation while maintaining fast spawn times comparable to system forks. The tool supports branching, allowing an agent to snapshot and resume its in-flight state to explore divergent reasoning paths. It includes support for Python, full TCP networking, and multi-tenant resource management.

forkd at a glance

Best for: Software engineer
Runs on: Linux, API
Pricing: Free
Released: May 2026
Category: AI coding tool
Website: github.com/deeplethe/forkd

Key features

KVM-based hardware isolation per child
Copy-on-write memory snapshots for fast spawning
Support for multi-vCPU and full TCP networking
Independent network namespaces and cgroup v2 memory limits
REST API for daemon control
Branch capability for mid-thought agent snapshots
Support for layered diff-snapshot chains

Use it for

AI agent fan-out for tool use and evaluation
Concurrent code interpretation and execution
Parallel agent reasoning with divergent tool or prompt inputs
Shared filesystem state across multiple sandboxed processes

Alternatives to forkd

Forked — It versions structured state with git-like branches and manages side effects safely.
Daytona — It provisions sandboxes in under 90 milliseconds with full security isolation.
CubeSandbox — Provides sub-60ms cold starts and hardware-level microVM kernel isolation.
Agent Sandbox — Provides a secure and isolated execution layer for autonomous AI agents.
microsandbox — Provides hardware-level isolation without needing daemons or complex setup.
Sandkiln — Runs untrusted code in hardware-isolated Firecracker microVMs.
AgentOS SDK — Runs agents directly inside the backend using WebAssembly and V8 isolates.
Sandcastle — Orchestrates AI coding agents in isolated Docker, Podman, or Vercel sandboxes.
AgentENV — Scales massive Firecracker environments across machines efficiently.
AgentFS — Provides isolated filesystem access and instant snapshots via SQLite.
Eve — Builds durable AI agents with file system based configuration and isolated sandboxes.
LayerFS — Mounts complete workspaces and commits tool calls instantly.

All forkd alternatives