XBTify

Project Name

XBTify

DESCRIPTION OF THE PROJECT

XBTify is a Farcaster mini-app that lets you create an AI agent capable of replying to casts or DMs where you’re tagged. The agent is trained on your own posts, learning your tone, knowledge, and style to act as your digital twin.

TELEGRAM USER NAME

• tg: @casocasocaso
• tg: @bianc8_eth
• tg: @itsmide_eth
• tg: @silver22_2k

REPOSITORY WITH THE PROJECT'S CODE

• AI Agent Server: https://github.com/builders-garden/xbtify-ai-service
• XBTify MiniApp: https://github.com/builders-garden/xbtify
• XBTify XMTP Base: https://github.com/builders-garden/xbtify-xmtp-app-base
• XBTify XMTP Agent: https://github.com/builders-garden/xbtify-xmtp-agent

VIDEO DEMO

https://youtu.be/CXoyZadUWh8

WEBSITE URLs

xbtify.me

BOUNTIES

BASE - MINI-APP SOCIAL – Leveraging the Farcaster social graph

We built XBTify as a Farcaster mini-app that allows you to clone yourself and get your AI twin.
When you're tagged in a cast, your XBT automatically generates context-aware replies that mimic your tone and perspective.
The social graph and user casts history is central: it is the main knowledge base of your XBT (see below for the AI part). 

XMTP – Best Use of the Agent SDK

We integrated the XMTP Agent SDK to make every XBT agent chat-ready and create new XBT agents via NLP.

Chat with every XBT agent

Users can chat with an XBT bot of their choice directly via XMTP DMs and have a 1-to-1 chat interaction.

Create new agents with xbtify.base.eth

Users can chat with xbtify.base.eth to create their agents via XMTP agent sdk, handling the alternative creation flow, parallel to the miniapp, with payments using wallet sendCalls of xmtp.

AI<>WEB3 TRACK

XBTify uses the publicly available data and interactions that Farcaster offers, which are fully owned and public (sufficiently) on-chain. This material is used as training material for the XBT AI twin.

The AI Agent, powered by GPT-5-mini, begins by analyzing a big creator's casts history to extract the distinctive vocabulary and the recurring themes of their content.
It then studies the creator’s reply history to capture the essence of their communication style, including tone, syntax, and typical response patterns for each identified topic. 
All these insights are then merged into a personalized style prompt, which guides the AI assistant when generating replies.

In parallel, a text-embedding model encodes the creator's casts history into a vector database. Through a Retrieval-Augmented Generation (RAG) pipeline, the agent retrieves the most relevant contextual information from this knowledge base to craft authentic, context-aware responses. Finally, the retrieved content, the style prompt, and the user’s query are combined to produce the final, personalized reply.