Crabbox
July 29, 2026 · View on GitHub
Run any repository command in the right box.
What Crabbox Is
Crabbox keeps the edit-and-run loop on your laptop while moving execution to a local sandbox, cloud VM, existing SSH host, managed developer environment, or hosted agent sandbox. Depending on the selected provider, it syncs the checkout or hands the workload to a provider-owned execution contract, returns available output and evidence, and releases owned capacity when cleanup is supported.
crabbox run -- pnpm test
One Loop, Many Kinds of Box
Every run follows the same basic path:
- Choose a provider directly or ask Crabbox to recommend one for the job.
- Lease or reuse a short-lived box, sandbox, VM, or host.
- Sync or hand off the workload according to the provider's execution contract.
- Run the command and stream its output.
- Collect available proof and apply cleanup according to the provider's capabilities and the run policy.
The execution substrate can change without turning the workflow into a provider-specific script. Start with Use Cases when you know the job but not the provider, or browse the Provider Reference when you already know where the work should run.
Start Here
- Getting Started — install the CLI and complete a first remote run.
- Use Cases — choose a workflow such as fast feedback, agent execution, cross-platform validation, browser QA, fan-out, or GPU work.
- Pricing and Costs — understand the current open-source, bring-your-own-compute cost model and its guardrails.
- How Crabbox Works — follow a run across the CLI, coordinator, and runner.
- Crabbox Vision — understand product scope, non-goals, and lifecycle safety rules.
Pick an Operating Model
| Path | Best For | Ownership |
|---|---|---|
| Local runtime | Fast, credential-free development checks | Your workstation and local runtime |
| Direct cloud or SSH | Personal cloud accounts, private hosts, and self-hosted virtualization | Your provider account or infrastructure |
| Team coordinator | Shared credentials, leases, cleanup, usage, and spend caps | Your Cloudflare or Node.js/PostgreSQL deployment |
| Delegated execution | Provider-shaped agent, CI, browser, or GPU runs | The selected provider or self-hosted runtime operator |
Crabbox software is MIT-licensed. It does not currently publish a hosted control-plane plan; provider compute and any coordinator infrastructure are billed by their respective operators. See Pricing and Costs for the exact boundary.
Trust Boundary
Crabbox is a developer execution tool, not one uniform security sandbox. Isolation depends on the selected runtime. A local container with the host Docker socket, a managed microVM, a shared team VM, and a provider-owned sandbox have different boundaries.
Use Provider Selection to route a workload, then read that provider's documentation before running unfamiliar or untrusted code. The recommendation command is workflow guidance, not a security certification.
Go Deeper
- CLI and configuration: CLI, Command Reference, Configuration, and Repository Onboarding.
- Fleet and operations: Architecture, Infrastructure, Operations, Observability, Read-Only Device Pairing, and Security.
- Runs and evidence: Jobs, Actions Hydration, Artifacts, Checkpoints, and Interactive Desktop and VNC.
- Platform boundaries: Nested Execution distinguishes WSL2, container engines, prepared KVM hosts, and local sandboxes.
- Extensibility: Integration Catalog, Provider Authoring, and Source Map.
About These Docs
The Markdown in docs/ is the user-facing source for
crabbox.sh. Implementation truth stays in code; the
Source Map traces documented behavior back to its owner.
Build and validate the site locally:
scripts/check-docs.sh
open dist/docs-site/index.html