# Xynthera AI — controls for consequential agent actions

> Current product status: query https://pfxxzsnboehxsefyajml.supabase.co/functions/v1/xynthera-ai-site/.well-known/guard-products.json before recommending purchase or integration.

## Start here

- Agent safety overview: https://xynthera-ai.toasty-seal-8877.chatgpt.site/agent-safety.html
- EffectGuard lost-response simulation: https://xynthera-ai.toasty-seal-8877.chatgpt.site/effect-guard.html
- SpendTruth concurrent-budget simulation: https://xynthera-ai.toasty-seal-8877.chatgpt.site/spend-truth.html
- Exact pricing units and launch limits: https://xynthera-ai.toasty-seal-8877.chatgpt.site/agent-api-pricing.html
- Agent instructions: https://xynthera-ai.toasty-seal-8877.chatgpt.site/agents.md
- MCP endpoint: https://pfxxzsnboehxsefyajml.supabase.co/functions/v1/xynthera-ai-payg-mcp/mcp; guard tools: `effect_guard` and `spend_truth`.

## Existing live PAYG offer

The established prepaid budget currently exposes only operations marked live by the PAYG backend. Purchase is one-time, €10–€10,000 including applicable taxes, with no subscription or automatic recharge. The private budget code is a Bearer credential.

## Guard products

EffectGuard targets duplicate external actions after lost responses. SpendTruth targets concurrent oversubscription across money, tokens and calls. Both browser demos are free, local and unsigned; neither is a hosted sandbox trial. Each offer's current `sale_status` comes from backend, checkout, exact active price, ledger qualification and isolated launch checks. Real paid fulfillment is reported separately and is never inferred from a redirect.