lamble.fun

Demo

About

Lamble is a meta-launchpad for people and AI agents. It brings together market intelligence and token-launch orchestration to help creators decide what to launch, where to launch it, and how to structure it. Rather than replacing existing launchpads, Lamble provides a single starting point above them. The project starts with research. Its web dashboard compares 19 launchpads across multiple chains using fees, revenue, indexed launch activity, completion data, and historical trends. A narrative explorer groups related tokens, identifies their launch venues, and charts their combined trading volume over time. This gives creators a way to investigate which ideas and launch mechanisms are attracting activity before choosing a venue. The launch workflow is conversational. Users describe their idea through Telegram, the web application, or an AI agent connected through MCP. Lamble turns that intent into a reviewable proposal, compares compatible launch routes, and produces an itemized quote. Crucially, it preserves each launchpad’s native configuration rather than forcing every venue into the same generic form. Creator fees, holder rewards, initial purchases, allocations, and vesting are available according to the selected route’s capabilities. The execution layer implements Pump.fun on Solana and Clanker on Base, with direct funding. Planning, funding, and deployment are separate steps, and spending requires explicit user approval. Transaction verification, public launch pages, and creator profiles connect the original proposal to an inspectable outcome. The idea is simple: creating a token should not require starting from scratch on every launchpad. Lamble brings research, venue selection, configuration, and execution into one workflow.

How it's made

Lamble has two applications. The intelligence dashboard uses Next.js 16, React 19, strict TypeScript, Tailwind CSS, and customized shadcn components, with Lightweight Charts for historical data. The launch application uses React and Vite, backed by Hono on Cloudflare Workers. D1 stores launch state, proposals, approvals, and accounting records; R2 stores media; Cloudflare Workflows coordinates conversations and execution; and a recovery cron reconciles unfinished operations. The intelligence layer uses Frames MCP to discover and access data sources, including DefiLlama financial data and Codex GraphQL launchpad and token-market data. Research outputs are normalized into a validated snapshot consumed by the frontend, so rendering the website never triggers paid API calls. The pipeline preserves source evidence, measurement windows, and coverage limitations. Narrative charts aggregate hourly volume for identified token pools, while missing values remain unknown rather than becoming invented zeros. The current refresh process is agent-assisted, not a continuously running indexer. The launch backend separates language-model reasoning from transaction execution. An OpenAI-compatible model helps interpret intent and refine proposals, while typed application logic handles validation, quotes, approvals, and spending. Each launchpad adapter describes its capabilities, validates venue-specific settings, builds transactions, verifies their contents, and reconciles the result. Pump integration uses the Pump SDK and Solana libraries; Clanker integration uses the Clanker SDK and viem. This allows a shared conversational interface without losing the differences between launch mechanisms. One particularly hands-on challenge was getting the Solana dependencies to run inside Cloudflare’s Workers runtime. We replaced a native bigint-buffer dependency path with a pure-JavaScript implementation and configured Anchor’s browser-compatible entry behavior. We also found that Pump creation plus the initial buy exceeded the transaction-size limit in our implementation, so the execution engine represents them as dependent operations rather than incorrectly promising an atomic launch-and-buy.