Siri for Web3

Tensorkit transforms EVM smart contracts into intelligent, voice-interactive applications. by leveraging MCPs and state of the art ML Models and LLMs We make it possible for anyone — not just engineers — to use and build on-chain logic by talking to it.

Developers can convert smart contracts into Model Context Protocol (MCP) servers, facilitating standardized interactions between smart contracts and AI systems. This automation eliminates the need for custom coding for each integration, streamlining the development process.​

Vision: Our vision is to revolutionize the decentralized application ecosystem by enabling smart contracts to seamlessly interact with AI agents and users. By integrating TensorConnect (Our Client Kit), dApps can evolve from isolated protocols to dynamic platforms capable of real-time data processing and user engagement, unlocking new possibilities in automation, decision-making, and user experiences.​

⚙️ How it works

You log in. You create a project. You drop in your contract — from Etherscan or manually. Tensorkit fetches the ABI. It reads your contract’s shape. It generates a Model Control Protocol server — an AI wrapper for your smart contract.

We give you a preview link with an embedded MCP client, that you can use to interact with your smart contract. Each preview comes with an embedded smart wallet for a seamless UX.

TensorConnect (coming soon) will enable you embed your client into your own frontends, dApps and mobile apps so your users can easily interact with your smart contracts, while other AI agents will benefit from being able to interact with your MCP servers.

🧪 Tech

• Built with Node.js and TypeScript
• ABI → JSON-RPC → OpenAPI → MCP
• Uses Whisper for local voice inference
• Generates MCP servers dynamically and serves endpoints
• Generated MCP server code can also be downloaded
• Clients connect via REST or WebSocket
• All interactions are standardized with the Model Context Protocol

...

Our initial design included deploying an SSE (HTTP) server for your MCP Servers, but we hacked around it for now to serve it as a serverlet within our main app.

We also had to re-imagine voice chat, and didn't know how complex the voice orbs that dynamically react to voice (Siri, chatGPT) are to engineer, but we figured out our way with Threejs and react, after so many trials and compatibility issues.

Our initial Previews (/preview) worked with rainbowkit & re-own. which is the ideal path for TensorConnect (Our client kit) ... However being able to trigger smart contract calls on the frontend, as a response from a server action is actually quite complex.

So turns out, integrating Privy, was an easier path, that actually provides a better UX on (/agent) path.

In the future we want to support both modes, client invocations (for backward compatibility) and Privy ( so all AI inference + smart contract calls) can happen on the server side.