Risk Alerts#
Risk alerts are how Ava earns its keep when nothing seems to be happening. Ava keeps watching in the background; you only hear from it when something matters.
What Ava watches#
- Aave health factor (HF). How close a borrow position is to liquidation — Ava warns you as it falls.
- Idle balances. Assets sitting in your wallet earning nothing — useful for catching incoming transfers that could be put to work.
- Token approvals. Unlimited approvals, and stale ones you no longer use.
- Wallet activity. Recent transactions — useful if you share a wallet or worry about compromise.
What an alert looks like#
⚠️ Your Aave position dropped to HF 1.3 — liquidation risk is rising.
Here's what's in your wallet: • 500 USDC • 0.2 WETH • 50 LINK.
Want me to help you top it up?
Three things to notice:
- Plain English first, numbers second. You know what's wrong before you read the math.
- The action proposal is at fire time, not config time. Ava shows what's actually in your wallet now and proposes a concrete fix that uses real balances, rather than a rule like "if HF < 1.3 → repay 300 USDC" set in advance. (Auto-mode health-factor protection is the other trade-off: fixed steps you set up once, so it can act without waiting for you.)
- An alert moves nothing. If you say yes, Ava prepares the transaction and you confirm it on the web. (Only Auto jobs you've already set up act without a prompt.)
Tuning alerts#
Alert settings in Telegram (/alerts) are coming soon. What you can control today, in Ava Studio:
- Monitoring on or off — the Ava Guardian switch on the Running page (
/run).
- Where alerts reach you — turn Email, Discord, Telegram or Slack notifications on or off on the Settings page (
/settings).
See Managing Notifications.
Alerts vs Actions#
Alerts are read-only: an alert is just a message. Actions you ask for need your Confirm on the web — in Manual mode you also sign in your own wallet; in Auto mode, they run under an on-chain agent permission with a total spending limit per token and an expiry. Standing Auto jobs like health-factor protection are the exception to the per-action Confirm: they act on a schedule with no per-run prompt, but still only inside that permission. See Alerts vs Actions →.
Edge cases#
- Timing. Health-factor protection in Auto mode checks hourly, so a fast market can move between checks.