Run Inference Guide¶

This guide explains how to run AI inference on WasiAI using the pay-per-use model.

Quick Start¶

1. Find a model in the catalog
2. Enter your input in the "Run Model" panel
3. Click "Run Model"
4. Sign the payment in your wallet (no gas!)
5. Get your result with transaction proof

Step-by-Step¶

1. Find a Model¶

Browse the model catalog to find an AI model:

• Search: Use keywords to find specific capabilities
• Filter: By category, price, or popularity
• Sort: By newest, most used, or price

2. View Model Details¶

Click a model card to see:

• Description: What the model does
• Pricing: Cost per inference
• Category: Model type
• Reputation: User feedback score
• Agent ID: On-chain identity

3. Prepare Your Input¶

In the "Run Model" panel:

1. Enter your input text or data
2. For classification models, you may add custom labels
3. Review the price shown

Example inputs by model type:

4. Execute Inference¶

Click "Run Model" to start:

1. First request: Sent without payment
2. 402 Response: Server returns price
3. Wallet popup: Sign the USDC authorization
4. Second request: Sent with payment
5. Result: Displayed with transaction link

5. Review Results¶

After inference completes, you'll see:

┌─────────────────────────────────────────┐
│ Result                                  │
│ ─────────────────────────────────────── │
│ Sentiment: Positive                     │
│ Confidence: 94.2%                       │
│                                         │
│ Payment                                 │
│ ─────────────────────────────────────── │
│ Amount: $0.0100 USDC                    │
│ Transaction: 0xabc123... [View]         │
│ Status: ✓ Verified                      │
│                                         │
│ Feedback                                │
│ ─────────────────────────────────────── │
│ Was this helpful? [👍] [👎]             │
└─────────────────────────────────────────┘

Understanding the Payment¶

What You're Signing¶

When you sign, you authorize: - Amount: Exact price shown (e.g., $0.01) - Recipient: Model's splitter contract - Validity: 60 seconds from signing - Token: USDC only

What You're NOT Signing¶

• No unlimited approvals
• No access to other tokens
• No permission to drain your wallet

Gasless Payments¶

You don't pay gas because: 1. You sign an authorization message (not a transaction) 2. A facilitator submits it to the blockchain 3. The facilitator pays the gas fee

Providing Feedback¶

After each inference, rate the result:

• 👍 Thumbs Up: Result was helpful/accurate
• 👎 Thumbs Down: Result was unhelpful/wrong

Why Feedback Matters¶

• Builds the model's reputation score
• Helps other users discover quality
• Incentivizes creators to maintain quality
• Recorded on-chain (ERC-8004)

Viewing History¶

Inference History¶

See all your past inferences:

1. Go to your Dashboard
2. Click Inference History
3. View: model, input, result, payment, timestamp

Transaction Verification¶

Every payment is on-chain:

1. Click the transaction link
2. View on Snowtrace
3. Verify: amount, sender, recipient, timestamp

Advanced Usage¶

Custom Labels (Classification)¶

For zero-shot classification models:

{
  "input": "How do I provide liquidity?",
  "labels": ["DeFi", "NFT", "Security", "Trading"]
}

Batch Processing¶

For multiple inferences, use the API directly:

const results = await Promise.all(
  inputs.map(input => runInference(modelId, input))
)

Programmatic Access¶

See the Inference API for: - Direct API calls - Integration examples - Response formats

Troubleshooting¶

"Insufficient USDC Balance"¶

Cause: Not enough USDC in your wallet

Solution: 1. Check your USDC balance 2. Get testnet USDC from faucet.circle.com

"Signature Rejected"¶

Cause: Wallet rejected the signing request

Solution: 1. Make sure you're on Avalanche Fuji network 2. Try disconnecting and reconnecting wallet 3. Clear browser cache

"Payment Expired"¶

Cause: Took too long to sign (>60 seconds)

Solution: Click "Run Model" again to generate a new payment

"Inference Timeout"¶

Cause: Model took too long to respond

Solution: 1. Try again (model may have been cold starting) 2. Use a smaller input 3. Try a different model

"Model Not Found"¶

Cause: Model ID doesn't exist or was delisted

Solution: Return to catalog and select a different model

Tips¶

Save Money¶

• Test with short inputs first
• Compare prices across similar models
• Consider a license for heavy use

Get Better Results¶

• Provide clear, specific inputs
• Follow the model's expected format
• Check the model description for tips

Verify Payments¶

• Always check the amount before signing
• Verify the transaction on Snowtrace
• Keep track of your spending in Dashboard