Workspace-scoped access
A credential is bound to one workspace. Requests for any other workspace are rejected outright.
VibeNAS security
Give your AI less of your machine.
Running AI tooling straight on your development laptop exposes far more than the project you're working on. VibeNAS puts that work in a purpose-built workspace with explicit limits on what it can reach and what it can run.
Current controls
A smaller blast radius.
AI-generated commands aren't treated as inherently trustworthy. Access is scoped, execution is brokered, and the genuinely dangerous capabilities simply aren't there to reach for.
Filesystem boundaries
Paths are validated before every file operation. Directory traversal and anything escaping the workspace is refused.
Separate execution identity
Developer execution runs as a workspace-specific Linux user, never as the VibeNAS platform account.
Brokered execution
Shared Developer commands pass through the execution broker instead of getting a shell on the host.
Network isolation
Shared execution has no outbound network and cannot reach platform-local services.
Bounded resources
Runtime, memory, process, CPU, and queue limits keep one workload from affecting anyone else.
Why isolate MCP?
Your laptop holds more than the project.
It also holds personal files, SSH credentials, cloud config, browser sessions, development secrets, local databases, and a dozen services that have nothing to do with the task you handed your AI.
VibeNAS moves the boundary: an isolated project workspace instead of broad access to the machine you use for everything else.
- Your personal home directory
- Your laptop's SSH keys
- Unrelated local projects
- Local browser sessions
- Outbound network, on shared compute
- An unrestricted shared shell
Capability boundaries
What your AI can actually do.
Capabilities follow the plan. Anything outside it is denied rather than quietly left open.
| Capability | Free | Developer |
|---|---|---|
| File access | Workspace only | Workspace only |
| Run code | No | Approved managed tools |
| Outbound Internet | No execution | Blocked |
| SSH | No | No |
| Git networking | No | No |
| Package installation | No | No |
| Background services | No | No |
| Exposed ports | No | No |
Auditable execution
Every managed command leaves a trail.
Execution requests get a request ID and record the request, start, completion, queue time, duration, executable, workspace, and an argument fingerprint — without writing raw command arguments into the audit event.
What we haven't verified yet
We publish security guarantees only after checking them against production. Encryption at rest, backup policy, retention periods, credential lifecycle, and account deletion behaviour are still being finalised.
A short list of verified controls beats a long list we can't substantiate.
Start with a workspace, not your whole computer.
Create a free VibeNAS workspace and give your AI controlled access to the files it actually needs.
Create a free workspace