Launching today
Hyperresearch
Durable, verified agentic search
25 followers
Durable, verified agentic search
25 followers
Hyperresearch does the research and shows the evidence. Ask a question, get a cited report; each cited claim is checked against its source, and all sources are kept in a searchable vault. On the DRACO benchmark, we are the world's number-one research agent. New: Answers MCP and API. Your agent calls ask and gets a short answer with word-for-word quotes, or a simple "not found" instead of a hallucination. We combine 4 AI search engines and 40+ other data sources for the widest and deepest access.
Free Options
Launch Team
Framer AI AgentsDesign and publish professional sites with AI
Promoted
Hi Product Hunt, I'm Jordan, I built Hyperresearch.
It all started as an open-source repo (still MIT). Every data API has blind spots because each provider sees a different slice of the web, so I overlap as many providers as I could wire in and one covers what another misses.
I was also annoyed that deep research wasn’t durable, so Hyperresearch keeps every source a run reads and you can search it later and build on your research instead of starting over.
What’s in it:
Research reports. Deep Search takes 3 to 5 hours across about 150 to 200 sources, and every cited claim is double-checked against its source before the report ships.
A source vault. The sources a run reads stay in your workspace, searchable by words or meaning, from the console or MCP.
Answers MCP, new this week. Add hyperresearch-answers to Claude Code, Cursor or any MCP client, and ask returns a short answer with word-for-word quotes checked against their pages, or not_found instead of a guess, while search returns ranked sources with the passages copied verbatim. Every account gets $5 of free credit, no card.
We score the top spot on the DRACO deep research benchmark.
On 100 BrowseComp questions (Answers MCP), Answers scored 89%, Parallel pro scored 73%, Perplexity Sonar Pro scored 43% and Exa answer scored 12%, meaning we're also the number one research task API out there.
Add a card and your first Deep run is free.
What I'd love from you: give Answers a lookup your agent gets wrong today, and tell me whether it answered or said not_found, and whether that call was right.
- Jordan
Keeping the sources around is the part that caught my eye. A follow-up question shouldn't mean finding the same pages all over again.