bouncerAI

BouncerAI is our answer to the chaos in token launches—no more bots, no more sniper attacks, and no more clueless speculators. Our “bouncer” uses AI to evaluate each user’s alignment with a project’s ethos, letting teams define an ideal user profile based on both vibe and knowledge. We built a “bouncer” system that acts like an AI-powered club doorman, letting projects define their ideal user profile. The bouncer asks short questions, checks a user’s vibe (are they aligned with the brand?), and verifies their knowledge (did they actually read the docs/whitepaper?). Only those who prove genuine interest, by answering voice-based questions and hitting a quality score threshold, receive a token allocation on our bonding curve. This means if you don’t truly understand the project and your vibe doesn’t match, you can’t buy tokens. By tapping into multiple AI agents running in parallel - one for knowledge, one for vibe - we generate a real-time score of user suitability. A separate “Question Agent” then refines follow-up questions to match the project’s style. We see this as a novel step for decentralized capital: merging AI gating with fair token economics. Instead of just open-season token sales, we’re offering a curated, community-first launchpad where real users - who actually care - get the lion’s share of participation. In short, BouncerAI brings focus and fairness to token launches, making them more about shared vision than mindless speculation.

TECH STACK Design: Prototyped and refined in Figma Frontend: Built with Next.js and deployed on Vercel Backend: Implemented in Node.js (TypeScript, Express, Redis) and deployed on a Droplet Smart Contracts: Developed with Foundry and deployed on Arbitrum Database: Managed via Supabase AI Agent Framework: LangChain, hosted on Autonome

FRONTEND Our frontend is built with Next.js. It was designed in Figma, then implemented with TailwindCSS for a responsive UI. Users authenticate via Privy, receiving JWT tokens that secure subsequent interactions with the backend. For visualizing token price data, we integrated the TradingView Charting Library, enabling real-time chart updates. Additionally, the bouncer avatar was designed in Spline to enhance user interaction. In the future, the avatar may dynamically react to user responses.

BACKEND The backend is built in TypeScript using Node.js and Express, with Redis for session management and Supabase for data storage. Session Handling User sessions are created in Redis after verifying Privy-provided cookies, ensuring secure authentication and state management. Agent Integration Once authenticated, user interactions (including project context) are forwarded to the AI agent component. These interactions are stored, maintaining a complete record of each user’s progress. Token Allocation and Signature Generation If a user passes the bouncer (meets the required knowledge and vibe criteria), the backend determines the token allocation. A secure signature (with a nonce to prevent reuse) is generated for users to purchase tokens up to their allocated limit. Price Tracking A continuous backend script monitors all active projects, fetching and updating their current prices in Supabase. The frontend then retrieves this data to render real-time price charts using TradingView, providing users with up-to-date insights into token performance.

DATABASE We use Supabase as our database to store: User metadata, Project data, Bouncer Configurations, Project interaction data, and Historical price data for every project contract.

SMART CONTRACTS We use a factory smart contract to deploy a unique token contract per project. These token contracts employ a linear bonding curve, with curve parameters set at deployment via the factory contract. Liquidity Management Each token contract holds both ETH and its corresponding tokens until the bonding curve reaches maturity. Transfers are restricted to the token contract itself, preventing secondary markets from bypassing the bouncer. Core Functions buy(): Allows token purchases on the bonding curve, up to an agent-set allocation, validated via a signature. sell(): Enables users to sell tokens back at any time on the bonding curve. deployLiquidity(): Once the curve matures, anyone can deploy liquidity (ETH plus tokens) into Uniswap, finalizing the tradable market.

AI AGENT FRAMEWORK Our AI bouncer uses LangChain, structured into two workflows: Interaction and Briefing, for user engagement and configuration. Interaction Workflow Parallel Evaluation Agents: Multiple evaluation agents can run in parallel. Currently, we use a Knowledge Agent and a Vibe Agent, though more can be added in the future. Each agent calculates a score and drafts a follow-up question. Subsequent Question Agent: After evaluation, a Question Agent refines the most relevant question (knowledge or vibe) based on the project's style. Adaptive Questioning: Typically 3 to 5 questions are asked, ending early if responses are weak. If the user’s score is sufficient, the system decides whether to ask a Knowledge or Vibe question next, ensuring a dynamic, context-aware flow. Briefing Workflow Project Setup: Users configure the bouncer by providing a text summary or whitepaper for knowledge checks, a list of descriptive adjectives for vibe checks, and bouncer attitude adjectives for the Question Agent’s tone and style. Prompt Engineering: Each agent’s system prompt is crafted to ensure user-defined configurations directly influence behavior, aligning the bouncer with the project’s brand, style, and difficulty. Briefing Agent (Future): A planned Briefing Agent will interact directly with users during project creation, providing a playground to fine-tune bouncer configurations.