CapaKit
from build to run
Last verified:
What is CapaKit?
CapaKit is a security tool that sandboxes the entire AI app lifecycle—from build to run—making agent-driven development safe and productive. Instead of only sandboxing runtime like most tools, CapaKit isolates the build phase where arbitrary scripts execute during npm install, secrets get baked into config, and dependencies install with full host access. It provides per-app policies with workload-level isolation, no inherited host environment, no broad filesystem access, and no network by default with outbound traffic requiring explicit allowance.
Key features include ephemeral single-use sandboxes for every build and run, secrets resolved on-demand instead of hardcoded, one-command flow to run AI app Kits straight from GitHub, structured testing from service to full integration tests run in isolation, and packaging AI apps as single .capakit archive files for sharing. CapaKit natively speaks MCP, A2A, and OpenAI-compatible APIs—the protocols agents already use. The tool supports standalone web interfaces with built-in UI and agentic integrations that plug into agents via MCP or skills.
CapaKit is for developers and teams going through AI-native transformation needing standardized practices, builders spinning up quick POCs with local/remote models and web UIs, and security-conscious engineers who don't want to run unknown or third-party code unsandboxed on their machine. The tool is currently macOS only and in public alpha, free to use with Apple Developer ID-signed and notarized releases.
CapaKit pricing
Pricing model: Freemium
Public Alpha, macOS only (for now), Free to use. CapaKit is currently free with no paid plans announced. The tool is in public alpha stage and available for anyone to use on macOS without cost.
CapaKit pros
- Sandboxes entire app lifecycle from build to run, not just runtime
- Per-app policies with workload-level isolation
- No inherited host environment or broad filesystem access
- No network by default—outbound traffic must be explicitly allowed
- Ephemeral single-use sandboxes for every build and run
- Secrets resolved on-demand instead of hardcoded into code
- One-command flow to run AI app Kits straight from GitHub
- Public alpha and free to use
- macOS releases are Apple Developer ID-signed and notarized
- Shell installer verifies CapaKit signature for security
- Structured testing from service to full integration tests in isolation
- Native support for MCP, A2A, and OpenAI-compatible APIs
- Package AI apps as single .capakit archive files for sharing
- Local-first—works even offline, you build and own it
- Agent-native—designed to be driven, written, and understood by coding agents
- Install as Codex skill with --global-skill flag
- Mount host files using --mount flag for controlled access
- Test in isolation with capakit test command
- Clone and customize any Kit from the registry
- Delta between exposed unsafe host work and secure sandboxed lifecycle
CapaKit cons
- macOS only—not available for Windows or Linux yet
- Public alpha stage—may have instability or missing features
- Ephemeral installed skill files deleted when command exits
- Runtime uses seatbelt sandboxes which may limit some operations
- Network access requires explicit allowance—extra setup needed
- No support for non-Mac operating systems currently
- Limited to ai apps with AI functionality—not general applications
- Requires coding agent to use capakit for creating Kits
- Model caching requires manual --mount setup
- Host network endpoints limited to local 127.0.0.1 addresses
Frequently asked questions about CapaKit
What is CapaKit?
CapaKit is a tool that securely builds and runs AI apps with coding agents by sandboxing the entire app lifecycle from build to run. It provides per-app policies with workload-level isolation, no inherited host environment, no broad filesystem access, no network by default, ephemeral single-use sandboxes, and secrets resolved on-demand instead of hardcoded.
What is an AI app Kit?
An AI app Kit is a self-contained, sharable unit that becomes a standardized AI app with CapaKit. It includes standalone web interfaces with built-in web UI or agentic integrations that plug directly into agents via MCP or skills. The Kit anatomy includes a capability manifest, workloads directory with source code, AGENTS.md with agent guidance, capability-test.yml for capability checks, and README.md.
How do I install CapaKit?
Install CapaKit with one command: curl -fsSL https://capakit.com/install.sh | sh. The macOS releases are Apple Developer ID-signed and notarized, and the shell installer verifies the CapaKit signature for security.
How do I run an AI app Kit?
Run an AI app Kit straight from GitHub with: capakit run https://github.com/capakit/hello-world-demo-kit. You can also install it locally as a Codex skill with --global-skill codex, or run a packaged .capakit file with capakit run hello-world-demo-kit.capakit.
What protocols does CapaKit support?
CapaKit natively speaks MCP (Model Context Protocol), A2A, and OpenAI-compatible APIs—the same protocols your agents already use. The same manifest runs locally today and is designed to run anywhere tomorrow.
How do I test an AI app Kit?
Run tests in isolation with capakit test. The command loads the manifest file, prepares the workload, invokes exec with preflight checks, and validates MCP tool responses. Tests run in managed mode with ephemeral seatbelt sandboxes on macOS.
How do I share an AI app Kit?
Package your Kit as a single file with: capakit kit package --out hello-world-demo-kit.capakit. You can also submit a PR to feature your AI app Kit in the official Registry and share it with the community at the View Kit Registry.
Who is CapaKit for?
CapaKit is for developers and teams going through AI-native transformation needing standardized practices, builders spinning up quick POCs with local/remote models and web UIs, and security-conscious engineers who don't want to run unknown or third-party code unsandboxed on their machine.
What platforms does CapaKit support?
CapaKit is currently macOS only for now. The public alpha version runs on macOS with Apple Developer ID-signed and notarized releases using ephemeral seatbelt sandboxes.
How does CapaKit handle secrets?
CapaKit resolves secrets on-demand instead of hardcoding them into code and configuration files. Secrets are kept in CapaKit secret stores, never baked directly into code, and are handled by default with minimal needs tailored for each app.