Clusy vs Elicit
Both call themselves research tools and both are, for opposite halves of research. Elicit works over the published record. Clusy works over your data and your compute.
At a glance
| Clusy | Elicit | |
|---|---|---|
| What it does | Reads the relevant literature, then plans and runs the experiment it implies | Searches, screens, and extracts structured evidence from published papers |
| Works over | Papers on arXiv and OpenAlex, your uploaded PDFs, and your own datasets and warehouse tables | A scholarly corpus it states at 138 million papers, plus PubMed and trial registries |
| Output | A notebook with executed cells, metrics, charts, and trained artifacts | Review tables, extractions, and reports designed to be auditable |
| Scholarly search | arXiv and OpenAlex, with citation-graph expansion, screening, and cited synthesis | A stated 138 million papers plus PubMed and trial registries, screened at review scale |
| Systematic review | Not a systematic-review product: no PRISMA workflow, no review-scale screening | The core product, with a PRISMA 2020 workflow built to be audited |
| Runs the experiment | Yes: reads the method, then writes and executes the code on GPUs | No |
| Compute | free 8 vCPU / 8 GB RAM CPU sandbox up to H100 / H200 GPUs (141 GB VRAM) | None you control |
The core difference
Elicit is an evidence workbench, and at evidence work it goes deeper than we do. It searches a corpus it states at 138 million papers alongside PubMed and trial registries, screens abstracts and full texts at a scale measured in tens of thousands, and extracts structured fields into a review table with the traceability a systematic review requires. Its published evaluation against Cochrane reviews is the kind of claim you can check, which is the right standard for a tool making assertions about evidence.
Clusy reads too, and then does something Elicit does not. The agent routes into a researcher profile when the task calls for it: it searches arXiv and OpenAlex, expands the citation graph through references and recommendations, screens what it finds, extracts structured evidence with citation spans, and returns a cited synthesis. It can parse the PDFs you upload, and it can clone a paper's reference implementation and read the actual source rather than the description of it.
Then it runs the thing. The same agent writes and executes notebook cells on managed compute, from a free CPU sandbox up to H100 and H200 GPUs, so the reported number in the paper becomes a measured number on your data, with branching to compare the paper's configuration against your variants. That is the split: Elicit is built to survey a literature exhaustively and defensibly, and we are built to read enough of it to run the experiment.
Choose Clusy if…
- You are reproducing or extending a method, not cataloguing one.
- The next step after reading is a training run, an evaluation, or a benchmark.
- You want the survey and the experiment done by the same agent, in one place.
- The deliverable is a measured result, not a review table.
Choose Elicit if…
- You are running a literature or systematic review.
- You need screening and extraction that an auditor could follow.
- The question is what the field has already reported.
- PRISMA-style rigour is a requirement of your output.
Frequently asked questions
- Does Clusy do literature review?
- It does scholarly research: the agent searches arXiv and OpenAlex, follows references and citations to expand a seed set, screens the sources, extracts structured evidence with citation spans, and returns a cited synthesis. What it is not is a systematic-review product. If you need PRISMA-compliant screening across tens of thousands of records with an audit trail a reviewer will accept, Elicit is built for that and we are not.
- Can Clusy read the PDFs I upload?
- Yes. The agent parses uploaded files and extracts clean text from PDFs and rendered pages, so a paper you supply becomes evidence it can screen and cite alongside what it finds itself.
- Can Clusy reproduce an experiment from a paper?
- Yes, and it is one of the most common uses. You describe the method and point at the data; the agent writes and runs the cells on managed compute, and branching lets you compare the paper's configuration against your variants side by side.
- Is there a workflow that uses both?
- Yes, and it is the right one for formal reviews: Elicit to survey and screen the literature exhaustively and produce the extraction table, then Clusy to implement the promising methods and measure them on your data. For less formal work, Clusy's own scholarly search usually covers the reading well enough to get to the experiment.
Sources
Claims about Elicit come from its own documentation, last checked : Elicit Systematic Review, PRISMA 2020 support, Evaluating Elicit's SLR. Quotas, hardware tiers, and pricing move; check the vendor before relying on a number, and tell us if we have something wrong.
See the agent do the work.
Free plan, no credit card. Describe an ML task and watch it get planned, executed, and reported in a notebook you control.
Try Clusy free