Many people start learning languages and would like to expand on their vocabulary, but struggle to stay motivated. It's especially difficulty when learning alone.

Without external accountability, it’s easy to lose momentum.

• Most goal setting app rely on self-reported progress, which is unverifiable and doesn’t drive true accountability.
• Many vocal building apps use static question banks that can be easily gamed — learners, especially younger kids, may choose to guess the answer or game the system without true mastery.
• Generic AI chatbots provide interesting conversation, but lack structured learning paths and measurable progress, and there's no real stake in the line to keep people motivated.

What if you could stake ETH to create a language learning goal, and an **AI agent tracks your progress **and have full control over whether you can get your stake back?

We introduce structured vocab goals: target number of words, difficulty level, and completion deadline. The agent will help suggest a goal for you to confirm.

You won’t just click buttons and there's no shortcuts. The AI will test you in a free-form quiz by:

1. Asking you to define the word
2. Asking you to use it in a sentence
3. Generating multiple choice options in real time for you to select from

AI will decide whether you’ve mastered the word and provide feedbacks, and the AI is the sole entity that may update your word progress on-chain.

You interact with the AI via XMTP protocol , each learning conversation happens via the messaging protocol. Behind the scenes, each conversation creates a dedicated wallet (via CDP agent kit), enabling the AI to update your learning progress, word by word. (The AI will ask you for some gas fund to get started)

If you complete your goal, you can claim your full stake back. If you fail, you can only claim a partial refund based on your progress, and the remaining stake will be sent to a public reward pool contract that can be used for future community learning events and incentives.

Leaners can also create a study group through our site — an XMTP group chat where a special group learning agent is auto-invited to make the learning experience more fun. The group agent is a separate agent and is still under development, and for now there is no on-chain component in the group chat, but it allows learners to engage together and stay motivated.

On-chain accountability: Learning progress is verified and enforced through on-chain state, updated only by the AI agent.

Dynamic AI quiz flow: The agent provides dynamic quizzes with free-form and ai generated quiz components — preventing simple gaming of the system.

Staking-based incentives: Learners stake ETH and can reclaim it only after demonstrating mastery — adding real financial motivation.

Transparent reward pool: Partial refunds from failed goals fund a public reward pool for community learning incentives.

Study group support: Users can create group chats with an auto-invited group agent for collaborative learning (early version).

Coinbase Agent Kit Integration: Each conversation has a dedicated CDP wallet — ensuring fine-grained control and security over on-chain updates.

Extensible architecture: The system is designed to allow future expansions (new languages, richer quiz types, group goal tracking, leaderboard features).

Website: https://xmtp-lexi-pal.vercel.app/ (Live On Base Mainnet)

LanguageGoal Contract: https://basescan.org/address/0x4Ac4525e441A034aD0D0Fe48A03AD4F95Fe9daF8

Individual Learning Agent XMTP Address: 0x0205918b99875a7b5ae7d3060f0ad4d9afcc4c4b

Group Chat Agent XMTP Address (under development):0x55fdc82920507ed1694a79b19c1025e7f12efac4

• XMTP protocol for messaging
• Coinbase AI agent kit & OpenAI API for the chat agent and agent wallet
• Next Js for Frontend (integrated with XMTP browser SDK)
• Solidity/Hardhat for smart contract
• Subgraph for indexing
• GCP for agent hosting & Vercel for frontend hosting
• Rainbowkit for Wallet Connection
• BaseName Integrated
• Live on Base Mainnet

• The project built for the Base Batch Messaging Hackthon, it was created from scratch entirely during the hackathon.
• Make use of the xmtp example repo: https://github.com/ephemeraHQ/xmtp-agent-examples as starter

npm install to install the relevant packages npm run dev to start the server

Code in the frontend3 folder

config/contract.ts Contains information related to contract

export const CONTRACT_ADDRESS = export const GRAPH_API_URL =

config/xmtp.ts Contains information on XMTP agent setup

defaultRecipient: defaultGroupRecipient:

yarn install to install the dependencies yarn start to get the agent running

Individual learning agent: code in xmtp-agent-examples/examples/xmtp-coinbase-agentkit folder Group chat agent: code in xmtp-agent-examples/examples/xmtp-coinbase-group folder

Here are some available commands you can use with the agent:

WALLET_KEY=
ENCRYPTION_KEY= <DB encryption Key>

OPENAI_API_KEY=
NETWORK_ID=base-mainnet
CDP_API_KEY_NAME= <ID for CDP API>
CDP_API_KEY_PRIVATE_KEY=
XMTP_ENV=production

LANGUAGE_LEARNING_CONTRACT_ADDRESS=<Language Learning Goal Contract>

Code in the contract folder

Compile: npx hardhat compile Deploy: npx hardhat run scripts/deploy.js --network base

SIGNER_ADDRESS=<Individual Learning Agent Address>

• We've integrated XMTP in our own frontend leveraging the XMTP browser SDK, for best accessibility and customization. This will enable to to define custom content type for messages in the future and customize display.
• AI holds you accountable: Only the AI agent decides if you’ve mastered the words and whether your stake is returned.
• On-chain vocab progress: It's has been seamlessly integrated into the chat flow and will be displayed alongside with the chat.
• Basename integration: Basename are fully integrated into the chat interface
• XMTP Group Chat integration
• Built with Security in Mind: Logic checks are performed before handing off data to the agent.

• Parents can stake ETH and set up a learning goal for their child , this will help helping onboard younger generations on-chain, giving them a fun and structured way to earn rewards through real effort.

• Learners can use it as a personal goal-setting tool — putting ETH at stake for their own vocab goals and letting the AI agent hold them accountable.

• Language teachers or tutors can set up private group chats with their students and integrate the AI agent as an additional practice tool.