Transparency
The whole point of Tessera is that nothing is hidden. Live protocol numbers, every liquidation, and every action the agent has taken — all on-chain, all here.
What's real vs. what's mocked — and why
This is a testnet build. Here's exactly where the line is, stated up front.
Real & verifiable
- The Stylus vault, its rules, and every lend / borrow / repay / liquidate tx — live on Robinhood Chain (chain 46630).
- The AI agent's decision logic and its auto-repay / liquidation transactions (the Live Drill is the production path, not a script).
- The agent's safety boundary: it can only reduce your own debt from an allowance you control.
- The reproducible gap backtest below — its number is locked by CI; the build fails if it drifts.
Mocked for testnet (and the mainnet plan)
- Price oracle. A Chainlink-style mock kept fresh by a scheduled keeper. Mainnet swaps in real Chainlink + a deviation guard (already written) at one address.
- Tokens (USDC, tAAPL/tTSLA/tSPY). Testnet mocks with a public faucet — they carry no value. Mainnet uses real USDC and issuer-tokenized equities.
- The Live Drill's gap. We move the mock price to create the crash you choose. The agent's rescue that follows is real and unscripted; only the trigger is ours (that's the point of a drill).
Does the AI actually protect you?
In 9 modeled overnight / weekend / earnings gaps, regime-aware protection avoided liquidation in 3 of the 4 cases that liquidated an unprotected position (75%).
Liquidations avoided
3 / 4
Of baseline liquidations
75%
Avg extra HF buffer
+0.45
| Modeled gap | Unprotected | Tessera (regime-aware) |
|---|---|---|
| AAPL · weekend −18% | safe HF 1.03 | safe HF 1.39 |
| TSLA · weekend −28% | liquidated HF 0.94 | saved ✓ HF 1.22 |
| SPY · weekend −9% | safe HF 1.09 | safe HF 1.55 |
| AAPL · overnight −12% | safe HF 1.04 | safe HF 1.36 |
| TSLA · overnight −22% | liquidated HF 0.94 | saved ✓ HF 1.21 |
| TSLA · earnings −20% | safe HF 1.04 | safe HF 1.48 |
| AAPL · earnings −10% | safe HF 1.22 | safe HF 1.67 |
| SPY · crash open −33% | liquidated HF 0.82 | saved ✓ HF 1.14 |
| TSLA · severe −45% | liquidated HF 0.69 | liquidated HF 0.94 |
A modeled result (not historical protocol performance), run through the same deterministic regime engine the live agent uses. Protection is not magic — a severe enough gap still liquidates a protected position, but always from a higher buffer (less bad debt). Reproduce + verify the number in CI: pnpm --filter @tessera/web test gap-backtest.
Live protocol numbers
Read straight from the vault — refreshed every 15 seconds.
Protection track record
What the agent has done — and the one number that matters for a lender: bad debt.
Liquidations
Every liquidation on this vault, on-chain.
Loading…
Agent activity
Every alert and protective action the agent has taken.
Governance & principles
No token, ever. No airdrops, points, governance coin, or fee tiers. There is nothing to farm — only yield and credit.
Non-custodial. The smart contract holds funds, never Tessera. The agent acts only through permissioned entrypoints and the USDC allowances you sign — and can only ever reduce your debt.
Conservative by design. Per-asset LTVs of 40–60% and liquidation thresholds of 55–75% are sized to absorb overnight and weekend gaps.