STAN is an always-on AI agent that automatically tips Rumble creators using USDT. It stakes your capital in Aave V3 to earn yield 24/7, then fires conviction-weighted tips when stream momentum peaks — viewer surges, rant bursts, sentiment spikes, and milestone crossings.
Built on Tether WDK + Arbitrum One + Aave V3 + Safe ERC-4337.
- Deposit — Fan deposits USDT → staked in Aave V3 on Arbitrum One
- Earn — USDT earns ~4% APY 24/7 via Aave
- Watch — STAN monitors the Rumble stream event feed
- Score — Brain scores creator momentum (velocity + engagement + milestone proximity)
- Tip — When conviction threshold met: withdraw from Aave → send USDT to creator via Safe UserOp
@stan/core — Brain (conviction engine), WDK wallet, Aave lending, Safe smart account
@stan/api — Express server, SSE stream, REST endpoints, GODSEYE dashboard
WDK_SEED (BIP-39)
└─ m/44'/60'/0'/0/0 → WDK EOA (Aave lending, signing)
└─ Owner of Safe smart account (ERC-4337, EntryPoint v0.7)
└─ Tips execute as UserOperations via Pimlico bundler
git clone https://github.com/ronkenx9/stan-agent.git
cd stan-agent
cp .env.example .env # fill in your keys
pnpm install
pnpm --filter @stan/api devOpen http://localhost:3001 — the GODSEYE dashboard loads with live SSE feed.
| Variable | Required | Description |
|---|---|---|
WDK_SEED |
Yes | BIP-39 mnemonic (12 or 24 words) |
ARB_RPC_URL |
No | Arbitrum One RPC (default: public endpoint) |
CREATOR_ADDRESS |
No | Creator wallet on Arbitrum (tip recipient) |
PIMLICO_API_KEY |
No | Enables Safe smart account for tip execution |
GROQ_API_KEY |
No | Enables LLM tip rationale (llama3-8b) |
PORT |
No | Server port (default: 3001) |
Send USDT + a small amount of ETH (for gas) to the WDK EOA address shown at startup. Arbitrum One gas is ~$0.01 per transaction.
- Conviction Engine — Scores stream momentum (viewer velocity 45% + milestone proximity 35% + engagement density 20%) and adjusts tip size: LOW 0.5x, MEDIUM 1x, HIGH 1.5x, MAX 2x
- Tip Surge — The longer since last tip, the bigger the next one (up to 3x after 60 min idle)
- Aave Yield — Capital earns ~4% APY in Aave V3 while waiting to tip
- Safe Smart Account — Tips execute as ERC-4337 UserOperations, not raw EOA sends
- 5-Gate Safety — Daily spend cap, momentum threshold, conviction tier, cooldown period, yield guard
- LLM Rationale — Groq llama3-8b generates one-sentence tip reasoning per tip
- OpenClaw —
GET /skill.mdexposes the agent skill definition for composability
| Trigger | Description |
|---|---|
viewer_milestone |
Fires when viewer count crosses a threshold |
new_subscriber |
Matches a % of subscriber dollar value |
rant_burst |
Fires when rant total in 60s exceeds threshold |
match_rant |
Matches a % of individual rant amount |
sentiment_spike |
Fires when chat sentiment score exceeds threshold |
gifted_sub_wave |
Fires per gifted sub in a wave event |
| Method | Path | Description |
|---|---|---|
| GET | /stan/health |
Health check + wallet + smart account status |
| GET | /stan/wallet |
EOA + smart account addresses, USDT balance |
| GET | /stan/lending |
Aave position, surge multiplier, time since last tip |
| GET | /stan/tips |
Full tip history with tx hashes |
| GET | /stan/config |
Current config (triggers, thresholds, daily cap) |
| POST | /stan/configure |
Update config |
| POST | /stan/deposit |
Deposit USDT into Aave V3 |
| POST | /stan/simulate-milestone |
Fire a mock stream event (demo) |
| GET | /stan/events |
SSE stream — real-time agent activity feed |
| GET | /skill.md |
OpenClaw agent skill definition |
- Wallet: Tether WDK (
@tetherto/wdk-wallet-evm) - Lending: Aave V3 via WDK (
@tetherto/wdk-protocol-lending-aave-evm) - Smart Account: Safe 1.4.1 (ERC-4337, EntryPoint v0.7) via
permissionless.js - Bundler: Pimlico (Arbitrum One)
- Chain: Arbitrum One (Chain ID 42161)
- Token: USDT (
0xFd086bC7CD5C481DCC9C85ebE478A1C0b69FCbb9) - AI: Groq llama3-8b (open-weights LLM)
- Runtime: Node.js + Express + TypeScript ESM
Base architecture adapted from solagent-vault, an open-source multi-agent Solana vault. Wallet layer, trigger logic, and agent brain are original work built during this hackathon.
Apache 2.0 — see LICENSE.