
Context Agent Sandboxes
8.0#1 in sandbox softwareFrom $20 a monthFree planAndroidiOSWebAPI
Visit

| Fact | Context Agent Sandboxes | nono |
|---|---|---|
| Audit | Commands, files, connections, and permission decisions are recorded in an append-only audit log with the actor and decision maker. | Filesystem operations, network attempts, and command executions can be recorded in a SHA-256 Merkle-tree audit trail and exported as JSON. |
| Credential handling | Standing secrets are kept out of the sandbox; credentials are issued at the edge after authorization and expire with the action or run. | The site says credentials are injected at the boundary and never exposed to the agent. |
| Isolation | Sandboxes run as non-root container pods with capabilities dropped, a seccomp profile, and no cluster service-account token. | It uses Landlock on Linux and WSL2 and Seatbelt on macOS to enforce filesystem allow-lists that cannot be loosened after application. |
| Network controls | Administrators can choose open networking, a domain allowlist or workspace-only access, with outbound decisions logged. | It can block network access, allow specific domains, filter API endpoints, and route filtered traffic through an upstream proxy. |
| Purpose | Each agent task runs in its own sandbox on the customer's cluster under the permissions of the person who requested it. | nono is a kernel-enforced sandbox CLI and SDK for isolating AI agents, scripts, build tools, and other processes. |
No prices published.
It suits teams that need agent tasks to run in isolated environments with configurable network access and recorded permission decisions. Enterprise deployment is aimed at customers running sandboxes on their own Kubernetes cluster.
nono suits developers who want to define and audit boundaries around AI-agent tool calls on Linux, macOS, or Windows. Its policy profiles and developer interfaces may fit teams that want to manage those controls alongside code.