A blazingly fast, sandboxed code execution environment powered by Nanvix and the GitHub Copilot SDK. Code runs inside an air-gapped, hypervisor-isolated microvm with no access to the host machine — no filesystem, no network, no host processes. The hypervisor boundary guarantees full isolation.
For architecture and implementation details, see doc/design.md.
- Node.js 22.5+ (required for
node:sqliteused by the Copilot SDK) - GitHub CLI (
gh) installed - A GitHub Copilot subscription
- Linux with KVM support or Windows with Windows Hypervisor Platform (WHP) (for microvm sandbox)
The Copilot SDK authenticates through the GitHub CLI. Sign in and install the Copilot extension before running nanvix-copilot:
# Authenticate with GitHub
gh auth login
# Install the Copilot CLI extension
gh extension install github/gh-copilotThese commands work in both Bash and PowerShell.
# Install dependencies
npm install
# Download Nanvix sandbox and runtime binaries
npm run setup
# Build
npm run build
# Run with a prompt
npx nanvix-copilot "What OS are you running on?"
# Specify a model
npx nanvix-copilot --model gpt-4.1 "Solve the Tower of Hanoi for 4 disks"
# Specify a runtime
npx nanvix-copilot --runtime python "Print all environment variables"
npx nanvix-copilot --runtime javascript "List files on /"All
npmandnpxcommands work on both Linux (Bash) and Windows (PowerShell).
nanvix-copilot [options] <prompt>
Options:
--model <name> LLM model to use (default: gpt-4.1)
--runtime <python|javascript> Runtime to use (default: auto-detect from prompt)
--verbose Show detailed execution info
--trace Show generated code before sandbox execution
--perf Show performance timing for each step
--nanvix-home <path> Path to Nanvix binaries (default: ./nanvix)
--setup Download Nanvix binaries and exit
--help Show help
This project is a prototype. As such, we provide no guarantees that it will work and you are assuming any risks with using the code. We welcome comments and feedback. Please send any questions or comments to any of the following maintainers of the project:
By sending feedback, you are consenting that it may be used in the further development of this project.
This project is distributed under the MIT License.