#opg $OPG I was digging through the OpenGradient docs, and AlphaSense is actually a pretty clever concept. It's a framework that lets you wrap verifiable AI workflows into tools that AI agents can call directly .
Here are some of the pre-built ones they have:
· Volatility AlphaSense – constant volatility forecasts for any asset. Use it for AMM fee scaling, portfolio risk management, or adjusting LTV ratios in lending protocols .
· PriceForecast AlphaSense – spot return predictions using time-series models. Yield strategies can use this to improve risk-adjusted returns .
· Sybil AlphaSense – feeds it wallet addresses, it outputs which ones are likely Sybils based on transaction history .
· Markowitz AlphaSense – classic mean-variance optimization. Feed it historical prices, get optimal portfolio weights .
The workflow is fully verifiable — data access happens through TEE-secured nodes, model inference runs with ZKML or TEE, and the whole thing is cryptographically auditable .
And here's the other piece: the Model Hub stores models on Walrus (decentralized storage). Upload a model, it gets a Blob ID, and it's immediately available for inference . The docs mention they've migrated over 100 AI models off their old IPFS architecture .
Basically, you can take any ONNX model, slap it on the Model Hub, and create an AlphaSense tool for AI agents in minutes. The SDK even has helper functions to turn models into LangChain or Swarm-compatible tools .
Verifiable AI workflows as composable building blocks. That's the takeaway.
@OpenGradient
Here are some of the pre-built ones they have:
· Volatility AlphaSense – constant volatility forecasts for any asset. Use it for AMM fee scaling, portfolio risk management, or adjusting LTV ratios in lending protocols .
· PriceForecast AlphaSense – spot return predictions using time-series models. Yield strategies can use this to improve risk-adjusted returns .
· Sybil AlphaSense – feeds it wallet addresses, it outputs which ones are likely Sybils based on transaction history .
· Markowitz AlphaSense – classic mean-variance optimization. Feed it historical prices, get optimal portfolio weights .
The workflow is fully verifiable — data access happens through TEE-secured nodes, model inference runs with ZKML or TEE, and the whole thing is cryptographically auditable .
And here's the other piece: the Model Hub stores models on Walrus (decentralized storage). Upload a model, it gets a Blob ID, and it's immediately available for inference . The docs mention they've migrated over 100 AI models off their old IPFS architecture .
Basically, you can take any ONNX model, slap it on the Model Hub, and create an AlphaSense tool for AI agents in minutes. The SDK even has helper functions to turn models into LangChain or Swarm-compatible tools .
Verifiable AI workflows as composable building blocks. That's the takeaway.
@OpenGradient