Security

Your code and your keys stay yours

An AI that can write files and run commands needs clear limits. Here is what WASDEV stores, what the AI is allowed to do and what leaves your computer.

Keys in the Windows vault

API keys and server passwords are stored encrypted in Windows Credential Manager, not in plain text files. Exported settings never contain them.

We never see your AI account

Signing in to Claude Code, Codex, Gemini CLI or Antigravity is done by the CLI itself, with its own login. WASDEV only checks whether you are signed in.

Local first

Projects live on your disk. Local servers listen on your machine only, with protection against hostile web pages.

Nothing lost

Automatic saving, version history and crash recovery for documents and recordings. Template files are never modified.

Passwords behind a PIN

The browser's password manager is locked by a PIN of 4 to 8 digits that you choose.

Sell your work freely

The 252 bundled libraries use licences that let you sell what you build — MIT, Apache-2.0, BSD.

Encrypted team sessions

Team calls and Live Share are joined with a 6-digit code that cannot be guessed offline (SPAKE2), and every message is then encrypted with AES-256-GCM.

Verified server certificates

FTPS servers are remembered by their certificate fingerprint: if it changes, WASDEV refuses to connect before your password is sent. SSH host keys are checked the same way.

No telemetry in the browser

The built-in browser engine runs with telemetry, crash reports, SmartScreen and sync turned off.

The AI's limits

An AI with guard rails

These rules are enforced by the application, not by a polite request in a prompt.

Inside the project only

The built-in file tools refuse any path outside the open workspace.

The app itself is off-limits

No AI can read, list, search or write the WASDEV installation folder, even in autonomous mode. Files that hold WASDEV keys or tokens are never readable.

No outside applications

Commands that launch other programs or open documents outside WASDEV are refused, and a watcher closes any window an AI CLI tries to open.

You approve the changes

Turn autonomous mode off and each edit by Claude Code or an API provider opens as a diff with Accept and Refuse; each command asks first.

Each CLI keeps its own sandbox

Codex, Gemini CLI and Antigravity are started in their own restricted modes. Codex, for example, is read-only until you switch the “Write files” tools on, and then writes only inside the workspace.

Read-only work modes

Plan, Review code and Security modes cannot modify files. Design mode blocks code and terminal tools unless you say “code it”.

Document code in a sandbox

When the AI writes Python to build a Word, Excel, PowerPoint or PDF file, that code runs at Windows Low Integrity: it cannot write to your projects or your settings.

Keys stay out of other programs

Terminals, debuggers, notebooks and test runs do not inherit your AI provider keys; each AI CLI receives only its own key.

You approve every send

Sending an e-mail, publishing a post, pushing code or deploying a site always asks you first, even in autonomous mode.

Network

What WASDEV sends, exactly

  • Licence check: your e-mail, your licence key, an anonymous identifier of the PC (a hash, not the serial number), the app version and the Windows version. The reply is signed and verified on your PC.
  • Update check: a simple request for the list of versions, carrying only the WASDEV version number.
  • AI requests: your messages and the files the AI reads go directly from your PC to the provider you chose — Anthropic, Google, OpenAI and so on. They do not pass through our servers.
  • With Ollama or LM Studio, AI requests stay on your machine.
  • The licence server keeps: e-mail, a hash of the key, the PC identifier, version, Windows version, and first and last check dates.