deeplethe/forkd

Health v2 · maintenance only

Flags: young

Adoption not part of the score

2752 stars · 207 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

forkd is a microVM sandbox runtime for AI agent fan-out, built on Firecracker, that forks up to 100 isolated child VMs in ~100ms from a warmed parent snapshot using copy-on-write memory sharing. It also supports BRANCH, which pauses a running sandbox and snapshots its in-flight state in ~150ms so agents can fork mid-execution.

Use cases

• spawn 100 isolated sandboxes for parallel AI agent execution in milliseconds
• fork a live agent VM mid-task to explore multiple branches
• run untrusted agent code with per-child KVM isolation
• share a warmed runtime (Python deps, loaded ML model) across many VM children cheaply
• snapshot and resume agent state for checkpointing or replay

When to choose

• you need to fan out many isolated AI agent sandboxes with near-fork(2) spawn latency
• you want per-VM KVM isolation without cold-boot cost
• you need to branch a running agent's state mid-execution
• you run on Linux with KVM and want a Firecracker-based workflow

When to avoid

• you need VMs on macOS, Windows, or hosts without KVM
• you need full cold-boot VM lifecycle management rather than snapshot-based forking
• your workloads diverge heavily from the parent so copy-on-write savings vanish
• you need a managed cloud sandbox service rather than self-hosted infrastructure

Facets

service · maturity active

agent-framework concurrency security artificial-intelligence cloud-computing developer-tools security rust cli self-hosted microvm firecracker copy-on-write snapshot kvm agent-sandboxing fan-out virtualization sandboxing ai-agents linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="deeplethe/forkd")