Please report security issues privately through GitHub's private vulnerability reporting: https://github.com/stfurkan/aidekin/security/advisories/new

Do not open a public issue for a security report. I aim to acknowledge within a few days.

aidekin runs entirely in the visitor's browser. There is no backend, no API keys, and no server-side state, which removes whole classes of risk but also bounds what can be enforced:

• The system prompt, the configuration, and any knowledge.bin are delivered to the browser, so treat them as public. Never put secrets, credentials, or private data in them.
• Because the model runs locally, a determined visitor can inspect or alter the prompt and the model on their own machine. The system prompt is configuration, not a security boundary.
• The widget loads in a sandboxed cross-origin iframe and is driven only by the embedding page (and any data-allowed-origins) through origin-checked postMessage.
• Cross-origin isolation (COOP/COEP) is required for voice; see DEPLOY.md.

This is a 0.x project. Security fixes target the latest commit on main.