Skip to content

Publish a Model Guide

This guide walks you through the complete process of publishing an AI model on WasiAI using the 5-step Publish Wizard.


Before You Start

Requirements

  • Wallet: Connected with some AVAX for gas (~$0.10)
  • Metadata: Name, description, category prepared
  • Cover Image: 1200x630px recommended (PNG, JPG, or WebP)
  • Model Artifact: At least one file to upload to IPFS (weights, configs, documentation)
  • Inference Endpoint: Your AI model's API endpoint URL
  • Payment Wallet: Address where you'll receive micropayments

Optional

  • Legal Terms: Terms of service for your model

Step 1: Basic Information

Enter the essential details about your model.

Fields

Field Required Description
Name Yes Display name for your model
Slug Yes URL identifier (auto-generated, editable)
Short Summary Yes Brief description of what your model does
Cover Image Yes Visual representation (1200x630px recommended)
Business Category Yes Industry category (Finance, Healthcare, etc.)
Model Type Yes Type of AI model (Sentiment Analysis, Classification, etc.)
Technical Categories No Technical classification tags
Author Display Name No Your name or organization
Social Links No GitHub, Website, X, LinkedIn

Tips

  • Name: Be descriptive but concise. "Crypto Sentiment Analyzer" > "Model 1"
  • Slug: Auto-generated from name; you can edit it for SEO
  • Cover Image: Use a relevant, high-quality image

Step 2: Customer & Technical Details

Describe your model for potential users and provide technical specifications.

Customer Sheet Fields

Field Required Description
Value Proposition Yes In one sentence, what does this model do?
Customer Description No Detailed description for users
Industries No Target industries (Retail, Finance, etc.)
Use Cases No Specific use cases
Expected Impact No What results can users expect?
Inputs Description No What inputs the model accepts
Outputs Description No What outputs the model produces
Examples No Input/output examples
Known Limitations No Risks and limitations
Prohibited Uses No What the model should NOT be used for
Privacy Notes No Data handling information
Support No How to get help
Deploy Options No Where the model can run

Technical Specifications (Optional)

Field Description
Tasks NLP tasks the model performs
Modalities Input/output modalities (text, image, etc.)
Frameworks PyTorch, TensorFlow, etc.
Architecture Model architecture details
Runtime Requirements Python version, CUDA, etc.
Resource Requirements VRAM, RAM, CPU cores

Step 3: Artifacts & Inference Configuration

Upload model files and configure your inference endpoint.

Artifacts (Required)

You must upload at least one artifact with the primary-weights role.

Field Required Description
File Upload Yes Upload file to IPFS
Role Yes primary-weights, adapter, inference-code, etc.
Notes No Description of the artifact

Supported Files: - Model weights (.bin, .pt, .onnx, .safetensors, .gguf) - Configuration files (.json, .yaml) - Documentation (.md, .pdf) - Code samples (.py, .js) - Archives (.zip, .tar.gz)

Inference Configuration (Required)

Field Required Description
Inference Endpoint Yes Your model's API endpoint URL
Payment Wallet Yes Address to receive USDC micropayments

Endpoint Requirements:

POST https://your-server.com/inference
Content-Type: application/json

Request: { "input": "user input here" }
Response: { "result": { ... } }

File Size Limits

File Type Max Size
Single file 100MB
Total per model 500MB

Step 4: Pricing

Pricing Options

Type Description When to Use
Per-Inference Price per API call Always (required)
Perpetual License One-time purchase For power users

Setting Prices

Per-Inference Price

Required. Users pay this for each API call.

Model Complexity Suggested Price
Simple (classification) $0.001 - $0.01
Medium (analysis) $0.01 - $0.05
Complex (generation) $0.05 - $0.20

Perpetual License Price

Optional. One-time payment for unlimited access.

Formula: Per-Inference × Expected Monthly Calls × 12-24 months

Example: $0.01 × 500 calls/month × 12 = $60

License Rights

Choose what each license grants:

  • API Access: Call the inference endpoint
  • Download Access: Download model artifacts

Step 5: Review & Publish

Review Checklist

Before publishing, verify:

  • [ ] Name and description are accurate
  • [ ] Cover image displays correctly
  • [ ] Pricing is competitive
  • [ ] Inference endpoint works (if custom)
  • [ ] All required fields are filled

Publication Process

  1. Click "Publish to Blockchain"
  2. Review the transaction in your wallet
  3. Confirm the transaction
  4. Wait for confirmation (~1-2 seconds)

What Happens On-Chain

When you publish:

  1. Model registered in MarketplaceV3
  2. Agent created in AgentRegistryV2 (ERC-8004)
  3. Splitter deployed via SplitterFactory
  4. Metadata stored on IPFS

Gas Costs

Network Estimated Cost
Avalanche Fuji ~$0.05-0.15
Avalanche Mainnet ~$0.10-0.30

After Publishing

Verify Your Listing

  1. Go to the model catalog
  2. Find your model (may take 1-2 minutes to index)
  3. Check all information displays correctly
  4. Test the "Run Model" functionality

Monitor Performance

Track your model's success:

  • Inference count: How many calls
  • Revenue: Total earnings
  • Reputation: User feedback scores

Update Your Model

To publish a new version:

  1. Go to your model's page
  2. Click "Publish Update"
  3. Use the same slug
  4. New version joins your model family

Troubleshooting

"Transaction Failed"

  • Check AVAX balance for gas
  • Ensure all required fields are filled
  • Try refreshing and reconnecting wallet

"Model Not Appearing"

  • Wait 1-2 minutes for indexing
  • Refresh the catalog page
  • Check transaction on Snowtrace

"Inference Not Working"

  • Verify endpoint URL is correct
  • Check endpoint is publicly accessible
  • Test endpoint directly with curl

"Image Not Loading"

  • Ensure image is under 5MB
  • Use supported format (PNG, JPG, WebP)
  • Try re-uploading

Best Practices

Metadata

  • Write clear, specific descriptions
  • Include example inputs/outputs
  • List any limitations or requirements
  • Add relevant keywords for search

Pricing

  • Research competitor pricing
  • Start lower to build reputation
  • Adjust based on demand

Maintenance

  • Monitor user feedback
  • Fix issues promptly
  • Update regularly
  • Engage with users