Evoke
Ask a research question and Evoke searches the academic literature, screens papers for relevance, pulls the findings into a table, and writes a cited report, showing its progress as it goes.
Period
- May 2026
Role
- Sole engineer
Type
- Site paused
Tools
What it does
I used Elicit a lot for class and wanted to see how it works under the hood, so I built my own version. You type a question and Evoke expands it into search queries, pulls papers from Semantic Scholar or CORE, screens each abstract against an inclusion rule you can edit, extracts the columns you care about into a table, and writes a report with citations. Progress streams to the page live, so you can watch each paper get screened.
There's also a lab section for systematic reviews, CSV export, a chat over your results, and PDF or PowerPoint export. An agent mode lets the model run its own search loop, following references until it decides it has enough.
Choices
Each stage uses a different model sized to the job: Haiku for screening because it runs on every paper, Sonnet for extraction, and Opus only for the final report. A full run costs about 15 to 25 cents. I used forced tool calls instead of asking for JSON in the prompt so the output always parses. The demo was bring your own key, with the key staying in your browser. The site is paused for now.
Limits
It only reads abstracts, not full papers, and screening is a single pass. Results are saved to a JSON file instead of a database. I planned an eval set but haven't built it, so I can't give you a number for how good the screening is.
