Show HN: Bubble Lab – Code-based agentic workflow platform (open-source) (github.com)

Bubble Lab is an open-source, AI-native workflow automation platform that compiles visual flows into production-ready TypeScript so developers fully own, debug, and deploy their automation code. It pairs a drag-and-drop "bubble" model (tools/nodes) with an AI assistant called Pearl that can generate workflows from natural-language prompts. Key features include one-click export to clean TypeScript, n8n JSON import, built-in execution tracing (logs, token usage, timing, memory), and templates (e.g., a ~50-line reddit-scraper) that demonstrate agentic chains like scrape → analyze (Google Gemini via OpenRouter) → structured JSON output. The project is Apache 2.0 licensed and runnable locally or via a hosted Bubble Studio; quick-start tooling (npx create-bubblelab-app) scaffolds TypeScript runtimes and examples. For the AI/ML community this matters because it emphasizes type safety, transparency, and portability in agentic workflows—addressing common pain points of opaque, vendor-locked builders. Technical implications: flows are first-class TypeScript with runtime packages (@bubblelab/bubble-core, bubble-runtime, ts-scope-manager, etc.), observability and error-handling are built in, and exports integrate with CI/CD and existing backends. Running the AI flow-generation features requires GOOGLE_API_KEY and OPENROUTER_API_KEY (otherwise manual flow building still works). The stack expects Bun for the API server, pnpm for monorepo management, and Node 18+, making it a practical option for teams wanting full control over agentic automation.