This demo gives an AI agent live access to enterprise data, so reports about it get a close look.

Don't open a public issue. Report it privately instead:

1. Go to this repository's Security tab.
2. Click Report a vulnerability (how private reporting works).
3. Describe the issue, how to reproduce it, and what an attacker could do.

Expect an acknowledgment within five business days. Fixes are merged to main.

Problems in this repository's code and configuration, such as:

• Reaching /chat or /tools on a reachable service without APP_API_KEY
• The chat page rendering attacker-controlled HTML or script
• Credentials leaking into logs, error messages, API responses, or the header
• A compose.yaml default that exposes a service or secret

• CData Connect AI itself, including its MCP server, LLM Gateway, and API. Report those to CData through Connect AI support.
• Your LLM provider or other third-party services.
• Prompt injection that stays within the agent's permissions. The agent can do whatever its Connect AI user can; limit that with read-only tools and the connection's per-user permissions, as the demo shows.

Revoke it first, then clean up:

• Connect AI PAT: delete it under Settings > Access Tokens in Connect AI, create a new one, and update CDATA_PAT in .env.
• LLM Gateway virtual key: revoke it on the Routing page's Virtual Keys tab and issue a new one. The provider key behind it stays private.
• LLM provider key: revoke it in your provider's console.

Treat a key that appeared in a screen recording or screenshot as leaked. Removing a secret from git history or blurring it in a video edit doesn't un-leak it. Revoke anything you published.