MarketIntel: Agentic Competitive Analysis

Gathering competitive intelligence is painful. You either spend hours manually browsing competitor websites, or you get an AI summary that hallucinates half the facts. Market Intelligence Agent solves this with a four-agent pipeline that runs autonomously. Paste in competitor URLs, hit start, and watch it work through a live SSE event stream. The scraper agent routes traffic through Bright Data's Web Unlocker, handling anti-bot protections, CAPTCHAs, and JS rendering walls. It crawls from the homepage to pricing, features, about, blog, and jobs pages. Each page gets cleaned, classified, and quality-scored before reaching the LLM. The analyzer agent runs structured extraction on every page, pulling out pricing plans with exact amounts, features grouped by category, team info, and news. Everything comes back as validated Pydantic models with prompt injection guards so adversarial text on competitor sites cannot hijack the extraction. The verifier agent cross-references every claim against the original scraped text in multiple passes. Claims without a direct source quote get flagged. Each finding has a confidence score and full citation chain. The reporter synthesizes everything into a report with comparison tables and recommendations, exportable as PDF, JSON, or CSV. Supports 20+ LLM providers via litellm, scheduled recurring analyses via APScheduler, SQLite persistence, and production guardrails including rate limiting, payload size limits, concurrent job caps, pipeline timeouts, and API key auth. The frontend is a Next.js 16 dashboard with real-time streaming, dark/light mode, and a pre-loaded Notion vs Obsidian demo so results are visible immediately. The live deployment uses Xiaomi MiMo 2.5 Pro through Opengateway (free but slower than commercial endpoints), keeping the entire stack at zero cost. Any litellm-supported provider can be swapped via env vars. Backend on Railway, frontend on Vercel.