BrowseAI-HQ/BrowseAI-Dev

Implements a multi-stage verification pipeline that decomposes compound claims into atomic facts, cross-references them across multiple search providers using hybrid keyword+semantic retrieval, and auto-calibrates confidence scores from real verification data rather than LLM self-assessment. Supports three depth modes—fast, thorough, and deep (agentic research with iterative think-search-extract cycles)—with Server-Sent Events streaming for real-time progress. Integrates with LangChain, MCP servers, and multiple search/LLM providers via pluggable authentication (Tavily, OpenRouter, custom keys).

Available on npm.