# Security, authority and privacy

Vuka is a personal AI workstation built on Kali Linux rolling. It inherits Kali's security-tooling focus and rolling update model; it is not a hardened appliance or a long-term-support distribution.

## Seven has full machine authority

Seven is intentionally not sandboxed. Its user-session process can view and control the X11 desktop, keyboard, pointer, applications, microphone/camera devices available to the account, files, network and supported serial/robotics interfaces. Vuka's separate `vuka-seven-rootd` service accepts requests from local `sudo`-group users over a mode-`0660` Unix socket and executes them as real UID 0 without a confirmation broker.

Each root request is written to the root-only `/var/log/vuka-seven-rootd.jsonl` audit log and the system journal. Logging provides accountability, not containment: unrestricted root can change the OS, data, logs and security controls. Compromise of Seven or an authorized local account must be treated as full-machine compromise.

## Local AI defaults

- Ollama runs as the dedicated unprivileged `ollama` account and binds to `127.0.0.1:11434` by default.
- No model weights are included or downloaded automatically. The user chooses a model and accepts its source, license, download size and storage cost.
- OpenCode is configured for the local loopback Ollama endpoint with sharing disabled. A user may separately configure cloud providers; Vuka does not include their credentials.
- The release audit found no API token, private key, personal Seven database, Wi-Fi credential or builder-local state in the ISO.

Vuka does not claim that all bundled upstream applications are telemetry-free. Review the privacy policy and network behavior of any browser, cloud model, plugin, model repository or third-party service before supplying personal or confidential data.

## Practical precautions

1. Use Vuka only on systems and data you are authorized to control.
2. Keep irreplaceable data backed up outside the machine and outside USB persistence.
3. Treat prompts, downloaded documents, repositories and model/tool output as potentially hostile because Seven can act on them with broad authority.
4. Keep Ollama and Seven loopback-only unless remote exposure is deliberately secured.
5. Review `/var/log/vuka-seven-rootd.jsonl` and the system journal after unexpected actions.
6. Use encrypted persistence or a full-disk encrypted install for sensitive data; unencrypted persistence is readable by anyone holding the USB.
7. Do not publish support bundles, logs, screenshots or Seven memory/state without reviewing them for secrets and personal information.

See `docs/SEVEN_INTEGRATION.md` for the exact two-plane authority design and `docs/KNOWN-LIMITATIONS.md` for unproven physical-device and Windows-application cases.