Open-source GPT Researcher updates deep research agent

gpt-researcher is an open-source deep research agent that uses planner and executor agents to break tasks into research questions, fetch from 20+ sources in parallel, and generate cited reports of 2,000+ words. This update introduces Jev context filtering: across a 28-task research benchmark, Jev retained relevant paragraphs at 73% versus 46% for embedding-based approaches, filtering in 1.7 seconds at about $0.115 per report. Without TYPESAFE_API_KEY set, it automatically falls back to local BM25 keyword ranking, requiring no API key or embedding model. It also adds MCP client integration, letting RETRIEVER=tavily,mcp search both the web and dedicated sources like GitHub; supports Google Gemini to generate inline report illustrations; and deep research mode takes about 5 minutes at roughly $0.4 per run. For research teams and developers using GPT Researcher, keyless local keyword filtering lowers the barrier to entry, while hybrid MCP retrieval brings internal databases and code repositories into research scope. For users needing long reports, Jev filtering cuts noise without extra cost, though unfiltered runs cover more ground at 65% higher cost.