AgentShield-X

AI assistant

Ask questions about AgentShield-X. Bring your own free Groq or Google Gemini key: the page detects which one it is and picks the best model your key can use.

Paste a key to begin. Free keys: console.groq.com or aistudio.google.com.

How this is kept safe

  • Your key stays with you. It lives in this tab's memory (or sessionStorage if you tick the box) and is sent only to the provider it belongs to, in a request header. This site has no server and cannot see it.
  • The page may contact only two hosts. A content security policy limits requests to api.groq.com, generativelanguage.googleapis.com and this site.
  • Answers are grounded in these docs. Excerpts are passed as reference data with an instruction not to follow them; excerpts that look like injected instructions are dropped.
  • Replies are shown as plain text, never as HTML, and credential-shaped strings are redacted.
  • Limits: 2000 characters per question, one question every 3 seconds, 30 per session, a 6-turn memory window.

What it cannot do

  • This browser guard is lighter than the full Python pipeline. For the complete input, retrieval and output defences use agentshield chat or the API route.
  • Model choice is a ranking heuristic over the list your key returns, not a benchmark. Free-tier quotas and model names change; if a model is rate limited the next best is tried.
  • It answers only from the project docs. It may still be wrong; check the linked docs.