Show HN: Open-Source Configurable AI Agents for Company Research (github.com)

Mira is an open-source, agentic AI system and npm-ready library for automating company data enrichment. It orchestrates specialized agents (discovery, internal pages, LinkedIn, Google Search, analysis) to gather configurable data points (industry, funding, headcount, recent news, etc.), attaches confidence scores (1–5) and source attribution, and emits real-time progress events. A Next.js frontend demo with Supabase-backed workspaces shows how to run research, view results, and generate AI-crafted outreach. Key conveniences include intelligent source selection (enable/disable crawling, LinkedIn, Google), composable configuration of data points and analysis rules, and smart early termination that stops processing once all targets reach high confidence—saving time and API costs. Technically, Mira is a TypeScript/Node.js v18+ monorepo using the OpenAI Agents SDK for multi-agent reasoning, ScrapingBee for scraping and Google Search, Zod for schema validation, Jest for tests, and Tailwind/shadcn UI for the frontend. It can be consumed as a library or run with the provided UI; required credentials include OPENAI_API_KEY and SCRAPING_BEE_API_KEY (plus Supabase for the demo). For ML/AI teams, Mira provides an extensible, reproducible pipeline for enrichment and scoring that’s easy to integrate into research, sales, or pipeline tooling—while also surfacing practical considerations around scraping credentials, cost control, and source trust via confidence metrics. Distributed under MIT.