nono
nono.sh
- Free plan
- Windows
- macOS
- Linux
- API
What nono is
nono is a free, open-source sandbox CLI and SDK for isolating AI agents, scripts, build tools, and other processes. It creates temporary sandboxes for individual tool calls, scoping files, network routes, arguments, and credentials. Landlock enforces boundaries on Linux and Windows/WSL2, while Seatbelt is used on macOS; the product says root is not required. JSON policy profiles can be version-controlled with code. SHA-256 snapshots record filesystem state before and after sessions, and sessions can be restored atomically. Operations are recorded in a Merkle-tree audit trail with a SHA-256 root for verification. Credentials are injected at the boundary, and proxy-mediated network rules can allow, deny, or require human approval. Interfaces include a CLI, Rust core library, and Go, TypeScript, and Python SDKs. It supports Linux, macOS, and Windows through WSL2, with 84% feature parity reported on WSL2.
Who it suits
It suits developers and teams isolating agent or process activity with self-hosted policies, credential protection, network controls, and audit trails.
Good to know, and what to watch
- Free and open source
- Kernel-enforced filesystem boundaries
- Scoped network and credential controls
- Version-controlled JSON policies
- CLI and multiple SDKs
- Native Windows is not supported
- WSL2 reports 84% feature coverage
- Some WSL2 advanced capabilities are unavailable
The verdict
nono provides self-hosted, kernel-enforced isolation for agent and process execution. Its policies, snapshots, credential handling, and audit trail address control and review, while Windows users must work through WSL2.
What the maker publishes
- Agent compatibility
- The site lists Claude Code, OpenCode, Codex, Antigravity, Goose, OpenClaw, GitHub Copilot, Qwen Code, Pi, and Hermes. source
- Agent integrations
- The site lists profiles or registry packages for Claude Code, OpenCode, Codex, Antigravity, Goose, OpenClaw, GitHub Copilot, Qwen Code, Pi, and Hermes. source
- Audience
- The maker describes use from individual laptops through fleets of agents and highlights composable policies, per-tool sandboxing, credential protection, and auditable execution. source
- Audit
- The audit trail records operations in a Merkle tree and commits a SHA-256 root at session end for integrity verification. source
- Audit
- Filesystem operations, network attempts, and command executions can be recorded in a SHA-256 Merkle-tree audit trail and exported as JSON. source
- Credential handling
- The site says credentials are injected at the boundary and never exposed to the agent. source
- Credential protection
- With proxy-based credential injection, the agent uses a localhost proxy and does not see the real API key. source
- Developer interfaces
- The documentation lists a CLI, a Rust core library, and Go, TypeScript, and Python SDKs. source
- Execution isolation
- It creates ephemeral, invocation-scoped micro sandboxes for tool calls, with scoped files, network routes, arguments, and credentials. source
- Installation
- The homepage offers a shell installer, Homebrew, and links to Debian, Fedora, Arch, and Nix installation instructions. source
- Instruction verification
- Its Sigstore-based provenance feature verifies whether agent instruction files were signed by trusted identities before execution. source
- Isolation
- It uses Landlock on Linux and WSL2 and Seatbelt on macOS to enforce filesystem allow-lists that cannot be loosened after application. source
- Kernel enforcement
- The OS sandbox uses Landlock on Linux and Windows/WSL2 and Seatbelt on macOS; the page says no root is required. source
- Network control
- The homepage describes proxy-mediated network access where policy can allow or deny requests and route out-of-policy actions for human approval. source
- Network controls
- It can block network access, allow specific domains, filter API endpoints, and route filtered traffic through an upstream proxy. source
- Policy profiles
- JSON profiles define agent access and can be version-controlled alongside code. source
- Pricing
- The maker's site and opened product pages provide no subscription price or paid plan; a maker-authored article describes nono as free and open source. source
- Product
- nono describes itself as agent runtime security with kernel-enforced boundaries for AI agents. source
- Purpose
- nono is a kernel-enforced sandbox CLI and SDK for isolating AI agents, scripts, build tools, and other processes. source
- Rollback
- SHA-256 content-addressed snapshots record filesystem state before and after sessions, and the page describes restoring a session atomically. source
- Rollback
- nono can take SHA-256 content-addressed snapshots before and after a session and restore the original state with a command. source
- SDKs
- The documentation describes a Rust core library and Go, TypeScript, and Python SDKs for integrating sandboxing into applications. source
- Support
- The documentation links to Discord for community support. source
- Support
- The documentation links to GitHub for source code and issues and Discord for community support. source
Checked against the maker's pages on 8 Oct 2026.