Launching today
Hyperresearch
Durable, verified agentic search
77 followers
Durable, verified agentic search
77 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
CloudflareDeploy serverless functions globally in seconds
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
The durability angle is what separates this for me — a 3-to-5-hour run across 150-200 sources is wasted if the evidence trail evaporates, so keeping every source in a searchable vault, with each cited claim checked against it, is the right call. One question on the vault: if a source page changes or disappears after a run, does a citation still resolve to the version that run actually read, or to the live page?
@jcbuilds Resolves to what the run read. But you can update the source afterward using our search API.
Thanks for the kind words!
The lookup ours gets wrong most is company facts, how many engineers and whether it got acquired, the data tools we tried were 6 to 18 months behind. Does an answer show how old the page it quotes is, so the agent can tell an old headcount from this year's?
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.
@oleksii_sekundant Exactly, that's the feature I've personally found the most useful.
Getting a plain not found instead of a confident hallucination is the promise I care about most here, and checking every cited claim against its source is what would make me trust a research agent with a real question.
@karimbenkeroum Glad you're resonating with it. Let me know if you give it a go!