Hypothesis App by Constructor Tech · Private Beta
Ask a research question and it returns the relevant papers, lays out the methods, claims, and contradictions across them, and helps you turn a direction into a testable, evidence-grounded hypothesis.
Upload your own papers and materials. Your collection is indexed into the same graph, so connections surface across your own library — and it stays yours.
See the white spaces in a field — open questions, thin evidence, and unresolved disagreements — and fill them with new hypotheses, grounded in and consistent with prior research.
Pressure-test a hypothesis and get a concrete validation plan — datasets, baselines, and metrics — each step traceable to the papers it came from.
Save hypotheses, gaps, and your own notes as artifacts — each one traceable to its source papers. Export them, reuse them across projects, and share by adding collaborators.
Hypothesis runs an MCP server, so your collections are searchable from MCP-compatible clients — Claude, Cursor, ChatGPT, or custom agents. Bring grounded, source-traceable literature search into the tools you already work in.
See where a field actually stands — which directions are crowded, which are open, and where the evidence disagrees — with the whole literature in view as you set your group's next project.
Get from a vague question to a focused, citable set of directions — and a testable hypothesis — without weeks of manual reading.
Stay current with a fast-moving literature and ground new bets in published evidence.
| Your research task | With today's AI tools | With Hypothesis |
|---|---|---|
| Reading across many papers | Chat summarizes one paper, or a few, at a time | Problems, methods, results, and concepts from many papers become explorable nodes in a single graph |
| Finding knowledge gaps | Ask for open problems and you get plausible but generic suggestions — not grounded in a specific body of work | Unexplored questions and under-connected areas surface across your collection and the literature — the gaps worth a new hypothesis |
| Trusting the output | Summaries often work from title and abstract; sources are hard to verify | Every result traces back to the source paper, built on full-text parsing |
| Forming a hypothesis | A plausible paragraph — ungrounded and hard to reproduce | Testable claims with datasets, baselines, and metrics — fully traceable to source |
Add your papers and the collection is indexed: each publication is parsed from full text into its methods and claims, then linked semantically to the rest.
State your research question in plain language — no query syntax, no keyword tuning.
Move through papers, methods, claims, and contradictions — filter, compare, and drill into any node back to its source.
Produce grounded hypotheses and a validation plan with datasets, baselines, and metrics — every artifact traceable to the papers it came from.
Your data is used only for your research. It is never used to train any models.
Your inputs, papers, and projects remain private — accessible only to you.
You keep full ownership of every artifact you generate — hypotheses, validation plans, and collections are yours.
Your data is hosted in line with applicable security and data-protection requirements.
The most valuable signal in science is a quiet contradiction between two results that are both solid. Hypothesis finds those, attributes them to specific claims and papers, and proposes what could explain both. No hallucination — only evidence with references.
Prof. Dr. Andrey Ustyuzhanin
Constructor University · ML, high-energy physics, AI for materials discovery
As a researcher, you often need to take birds-eye view of a field, rather than focusing on one paper at a time. Hypothesis maps the literature by revealing claims and contradictions, and methods, helping understanding and accelerating work.
Prof. Dr. Ivar Martin
CASE Senior Scientist, University of Chicago · condensed-matter theory
We're opening Hypothesis to a small group of active researchers. At this stage, feedback from researchers shapes the product directly — beta members have a real say in where it goes next.
| How access works | Join private beta → register → start your first exploration. |
| Who qualifies | Active researchers — PIs, postdocs, PhD students, and industry R&D. Seats are limited. |
Seats are limited and reviewed individually.
Yes. Beta access is free for invited researchers.
Yes. Upload your own papers and materials and they're indexed into the same graph — so connections surface across your own library. Your collection stays yours.
Those tools either rank papers by keyword or summarize them one at a time on top of a search index. Hypothesis builds a graph of how papers relate — their problems, methods, claims, and contradictions — and reasons over that structure. It works from full relational context and shows the source behind every conclusion.
Active researchers — PIs, postdocs, PhD students, and industry R&D. Register now, seats are limited.
Your data is never used to train our models, stays private to you and your invited collaborators, and is hosted under applicable security and data-protection requirements. You keep full ownership of everything you generate. (See Your data, your IP, your terms.)
Yes. Hypothesis runs an MCP server (Model Context Protocol), so you can connect it to MCP-compatible clients — Claude, Cursor, ChatGPT, or your own agents — and search your indexed collections from inside the tools you already use. Access is authenticated with a personal API key, and every result still traces back to its source paper.
Join our researcher community on Discord — beta feedback shapes the roadmap.
At this stage, feedback from researchers shapes the product directly. If something seems wrong, missing, or useful, we want to hear it — you'll have a direct line to the team building Hypothesis.
