An open source sandbox engine for AI agents that restricts what they can execute, read, write, and reach on the network, with default-deny policies.
nolabs-ai/nono
Kernel-enforced sandbox and security layer for AI agents, controlling their tools, credentials, network access, and file changes.
- Stars
- 4,876
- Forks
- 327
- Open issues
- 284
- Last commit
- 6 hours ago
What is it?
nono is an agent runtime security tool that runs AI agents such as Claude Code, Codex, Copilot, and Goose inside a kernel-enforced sandbox. Each tool the agent calls runs in its own ephemeral sandbox with explicit permissions for files, network, environment variables, and credentials, so the agent itself cannot reach secrets it was not granted.
Policies are defined in composable JSON profiles outside the agent's control. A profile can, for example, allow the aws CLI to use a credential while blocking curl from it, or restrict gh to reading issues and posting to a single organization. Network access is filtered by domain and by HTTP method and path. Skills, hooks, and instruction files can be verified with Sigstore signatures against a list of trusted publishers.
nono also keeps a tamper-evident audit record of sessions and security decisions, and can restore tracked files from verified snapshots after an agent makes a mistake. It runs on macOS, Linux, and Windows via WSL2 as a single binary without root privileges, and provides Rust and Python library bindings.
Key features
- 01Per-tool sandboxing with explicit file, network, environment, and credential permissions
- 02Credential injection through a proxy so API keys stay outside the agent's sandbox
- 03Domain filtering and HTTP method and path rules for network access
- 04Sigstore-based verification of skills, hooks, and instruction files
- 05Tamper-evident audit log of sessions and security decisions
- 06File change review and rollback to verified snapshots
Use cases
- Running coding agents on a developer laptop without giving them access to all credentials
- Enforcing command and network policies for agents inside CI workflows or containers
- Verifying that agent skills and hook scripts are signed by trusted publishers before use
- Keeping an auditable record of what an autonomous agent accessed and changed
Quick start
- 1
Install nono with Homebrew (macOS / Linux)
brew install nono - 2
Pull the Pi agent profile from the registry
nono pull nolabs-ai/pi - 3
Run the Pi agent inside the nono sandbox
nono run --profile nolabs-ai/pi -- pi
Things to know
- Windows is supported only through WSL2, not natively.
- The project is approaching 1.0, so APIs may still change.
Topics
FAQ
- What is nono?
- Kernel-enforced sandbox and security layer for AI agents, controlling their tools, credentials, network access, and file changes. nono is an agent runtime security tool that runs AI agents such as Claude Code, Codex, Copilot, and Goose inside a kernel-enforced sandbox. Each tool the agent calls runs in its own ephemeral sandbox with explicit permissions for files, network, environment variables, and credentials, so the agent itself cannot reach secrets it was not granted.
- Is nono free to use? What license does it have?
- nono is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is nono still maintained?
- Actively maintained. The last commit to nono was on Nov 6, 2026.
- What can I use nono for?
- Running coding agents on a developer laptop without giving them access to all credentials Enforcing command and network policies for agents inside CI workflows or containers Verifying that agent skills and hook scripts are signed by trusted publishers before use Keeping an auditable record of what an autonomous agent accessed and changed
- How do I install nono?
- Install nono with Homebrew (macOS / Linux): brew install nono Pull the Pi agent profile from the registry: nono pull nolabs-ai/pi Run the Pi agent inside the nono sandbox: nono run --profile nolabs-ai/pi -- pi
- Where can I get nono?
- The source code and installation instructions are on GitHub: https://github.com/nolabs-ai/nono
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Local-first runtime security layer that reviews AI agent shell commands, file access, package installs and MCP tool calls before they run.
Open detection-rule format for AI agent security threats, with executable, testable YAML rules for prompt injection, tool poisoning, and MCP attacks.
Policy enforcement, zero-trust identity, execution sandboxing, and SRE tooling for autonomous AI agents, usable from Python, TypeScript, .NET, Rust, and Go.
Zero-trust sandboxed runtime on Kubernetes that maps many stateful AI agents onto a small pool of workers by snapshotting idle actors.
Security scanner that checks AI agent skills (Claude Code, Codex, MCP) for prompt injection, data exfiltration and supply-chain risks before you install them.