Concave AI Sandbox is designed to execute untrusted AI code securely and efficiently.
This platform offers a self-hostable code execution sandbox, leveraging Firecracker microVMs. It boasts sub-200ms startup times, achieved through snapshot-based warm pooling.
About Concave AI Sandbox
Concave delivers secure, isolated environments for executing untrusted code. Its architecture is built upon Firecracker microVMs, gRPC orchestration, and contemporary infrastructure tools.
- Rapid VM provisioning using golden snapshots.
- A gRPC control plane for managing the VM lifecycle.
- A streaming data plane facilitating file transfers and real-time output.
- An HTTP API gateway equipped with authentication.
- A Python SDK for straightforward integration.
- A comprehensive dashboard and documentation site.
Getting Started
Refer to the SETUP.md guide for comprehensive deployment instructions.
- Establish a GCP project and domain.
- Configure Terraform and Ansible.
- Deploy the necessary infrastructure.
- Install the Python SDK using: pip install concave-sandbox.
Documentation
Contributing
Contributions to the project are encouraged. Refer to CONTRIBUTING.md for detailed guidelines.
Security
Should a security vulnerability be discovered, please create an issue. All reported security concerns will be addressed promptly.
License
The project operates under the MIT license.
