---
title: "Security at Alran Code: your code, your rules"
url: https://alran.dev/en/security
locale: en
updated: 2026-11-16
---

Security

# Your code. Your rules.

An agent that runs commands should ask first. Alran Code does, and it shows you everything it does.

## You stay in control. {#h-control}

### Permission modes

Ask, Auto or Full Access. In Auto, steps with effects outside your workspace wait for your OK.

[Read more](https://alran.dev/en/docs/plan-mode)

### Plan before change

Plan Mode reads and plans, but changes nothing until you say so.

[Read more](https://alran.dev/en/docs/plan-mode)

### Every step visible

Files read and written, commands and their output: all in the conversation, one keystroke away.

[Read more](https://alran.dev/en/docs/tui)

### Sessions stay local

Your sessions, settings and sign-in live in `~/.alran` on your machine.

[Read more](https://alran.dev/en/docs/config)

### No password in the terminal

You sign in in the browser or confirm a code. Signing out revokes the session at the platform at once.

[Read more](https://alran.dev/en/docs/auth)

### Your keys go direct

Calls with your own provider key go straight to the provider. Ottili doesn’t see, meter or bill them.

[Read more](https://alran.dev/en/docs/byok)

## Open, so you can check. {#h-open}

The whole agent is on GitHub under the MIT license. No usage is reported to any host unless you set a telemetry endpoint yourself.

[Read the code](https://github.com/Alran-AI/code)

- Sign-in session stored with owner-only permissions.

- Keys for your own providers can stay in environment variables, never on disk.

- MCP keys are read from the environment, not from the config file.

- Werk 1.1 is processed in the United States.

## Found a vulnerability? {#h-report}

Report it privately on GitHub: **Security → Advisories → Report a vulnerability** on the Alran Code repository. Or write to [support@ottili.one](mailto:support@ottili.one).

Include the version, your operating system and Node.js version, a minimal reproduction and the permission boundary you expected. Use test accounts and synthetic files; never test other people’s accounts. Permissions and the sandbox reduce risk, they don’t replace reviewing untrusted plugins, MCP servers and shell commands.
