← All projects
AI PipelinesOnline

Evoke

Open-source Elicit clone

A literature review is really four different jobs, and they don't all need the same intelligence. Evoke splits them apart and routes each stage to an appropriately-sized model — cheap and fast for screening volume, strongest for synthesis.

Screening is the stage researchers actually argue with, so the criteria are natural language and the re-screen loop is editable: change what counts as relevant and the corpus re-filters without starting over.

What it does

  • Four stages: search → screen → extract → synthesize
  • Searches a 200M-paper corpus via the Semantic Scholar API
  • Screens against natural-language criteria you can edit and re-run
  • Stage-appropriate model routing (Haiku → Sonnet → Opus)
  • Live SSE progress while the pipeline runs
  • Produces a cited narrative review, not a link dump
Next projectPipeline Guard