Unlike chatbots that answer from training data, this agent actively searches the web. Plans sub-questions, fetches sources, synthesizes a cited report. 1-3 min per query.
POST /api/agents/deep-research · Method: tasks/send or tasks/sendSubscribe · Agent card: GET /api/agents/deep-research
Regular research uses LLM training knowledge only. Deep research actually searches the web (DuckDuckGo), fetches pages, and cites them. Slower but more current and verifiable.
1-3 minutes typically. Plans 4-6 sub-questions, fetches 8-12 sources, synthesizes with citations.
Yes — pulled from actual fetched web pages. Each [N] marker references the URL list at the report's end.
Some sites block scraping or load slowly. Agent gracefully degrades to LLM knowledge with limitation flagged.