Fluent

Fluent is an AI-powered Web3 learning platform that transforms how people understand blockchain concepts. Instead of reading static documentation or watching endless YouTube tutorials, users learn by highlighting any Web3 sentence or concept directly in their browser and receiving instant, personalized explanations.

Architecture Overview Fluent combines a browser extension frontend, an AI-powered agent backend, and blockchain infrastructure to create a seamless learning experience.

Core Technologies & Integration

1. Browser Extension (JavaScript/Chrome Extension API)

Content scripts inject into any webpage to detect text selection

Captures highlighted sentences and sends them to the backend agent

Displays real-time explanations in a floating tooltip UI

Built with vanilla JavaScript for maximum compatibility

1. AI Agent Backend (Python + uAgents Framework)

uAgents: Fetch.ai's autonomous agent framework handles all backend logic

ASI:One Integration: Uses three specialized models:

asi1-mini: Core concept extraction and explanation generation

asi1-graph: Advanced graph reasoning and learning path optimization

asi1-agentic: Multi-step reasoning for complex queries

MeTTa Knowledge Graph: Symbolic reasoning engine that stores user's learning as queryable graph nodes and relationships

Runs as a mailbox agent, discoverable on Agentverse while hosting local REST endpoints

1. Real-Time Blockchain Data (Blockscout MCP)

Integrates Blockscout's Model Context Protocol for live on-chain examples

When explaining "relayers," shows actual recent relayer transactions on Base

Enriches static definitions with real transaction data, gas costs, and contract interactions

Uses REST API calls to https://mcp.blockscout.com/mcp for chain-agnostic data

1. Database & User Management (Supabase)

PostgreSQL database with real-time subscriptions

Stores captured sentences, user knowledge graphs, quiz results, and NFT metadata

Row-level security for multi-user support

Google OAuth integration for seamless authentication

1. Soulbound NFT System (Solidity + Base)

Custom ERC-721 contract deployed on Base Sepolia

Quiz-gated minting: users must score 80%+ to earn domain badges

Non-transferable tokens serve as verifiable learning credentials

Metadata stored on IPFS via Pinata, includes user's knowledge subgraph

1. Frontend Dashboard (Next.js + React)

Visualizes user's knowledge graph with interactive node-link diagrams

Domain-specific views (DeFi, L2s, DAOs) with progress tracking

Badge gallery showing earned NFT credentials

Real-time updates via Supabase subscriptions

Partner Technology Benefits ASI Alliance Stack (uAgents + ASI:One + MeTTa)

uAgents provided the perfect framework for building stateful, conversational AI that can handle both REST endpoints and agent-to-agent communication

ASI:One models were specifically designed for agentic workflows, making concept extraction and graph reasoning much more reliable than generic LLMs

MeTTa enabled true explainable AI—users can see exactly what the system knows and why, crucial for educational applications

Blockscout MCP

Transformed static explanations into live, contextual learning with real transaction examples

The standardized protocol made blockchain data integration trivial—just POST requests to get formatted, AI-ready data

Multi-chain support means explanations can reference actual activity across different networks

Base Integration

Low gas costs made badge minting practical for educational use cases

Fast transaction finality provides immediate feedback for quiz completion

Strong ecosystem support for NFT tooling and wallet integration

Notable Technical Innovations

1. Adaptive Explanation Engine The system maintains separate knowledge states for each user. When explaining new concepts, it queries the user's MeTTa graph to determine what they already know, then generates contextually appropriate explanations. A beginner gets "DeFi is decentralized finance," while someone who already knows DeFi basics gets "DeFi protocols like Aave use over-collateralization to manage lending risk."

2. Live Transaction Context Instead of static examples, every explanation can include real, recent blockchain activity. The agent queries Blockscout MCP with extracted concepts and surfaces relevant transactions, making abstract concepts tangible.

3. Quiz-Gated NFT Minting The smart contract integration was particularly elegant—quiz questions are generated by ASI:One based on the user's captured knowledge graph, ensuring personalized difficulty. Only after proving understanding (80%+ score) can users mint domain-specific badges.

4. Cross-Extension Learning Persistence The knowledge graph persists across browser sessions and websites. A user can learn about "yield farming" on one site, then see that knowledge referenced when they encounter "liquidity mining" elsewhere.

5. Agent-Extension Communication The browser extension communicates with a local agent that's simultaneously registered on Agentverse, enabling both private learning and potential collaborative features with other agents in the ecosystem.