Living literature reviews · auto-research
Subscribe to a research question. Perennial writes a structured, citation-verified survey — then ingests each week's papers and tells you which of its own claims are now supported, contradicted, or superseded.
Maintenance, not search
Everyone else runs a search. Perennial keeps the answer alive.
arXiv adds roughly 28,000 papers a month, and "caught up" never arrives. One-shot tools hand you a survey that's outdated the day you read it. Perennial maintains the survey week after week — so it gets more valuable the longer you keep it.
New work from arXiv, OpenReview, bioRxiv, PubMed and Semantic Scholar is ingested, deduplicated, and integrated into only the affected sections.
Every reference in a candidate paper is checked against Crossref and arXiv. Fabricated and citation-mill sources are filtered before they ever reach your survey.
Every claim is pinned to a paper, section, and quoted span. Diff any two versions, trace provenance, and export verified BibTeX straight into your manuscript.
What Perennial does
Claim-level entailment applied to a living artifact — not a retrospective per-paper lookup, and not a chatbot's one-shot guess.
Seed a research question — or hand over your own related-work draft — and Perennial generates a structured survey where every claim is pinned to a paper, section, and quoted span.
Fabricated citations rose roughly tenfold since 2023. Perennial verifies each reference against Crossref and arXiv and quarantines anything that doesn't resolve.
No "nothing changed" noise. When new results actually touch your survey, Perennial integrates them into the affected sections and sends a plain-language digest of what shifted — and what now disagrees with a claim you cite.
Point Perennial at results you've published. If a new paper contradicts or supersedes them, you hear it from your survey first — not from a reviewer.
The living survey isn't a walled garden. Drop a section into your related-work, pull the verified BibTeX into your manuscript, or share the canonical review with your lab so it becomes everyone's reference infrastructure. Per-claim provenance travels with every export.
Get started
Describe the research question — or upload your own related-work draft. Perennial builds the first structured, citation-verified survey, claim by claim.
Every week, new work is ingested across the major sources, screened for fabricated citations and AI-slop, and integrated only where it actually changes your survey.
A short email tells you which claims are now supported, contradicted, or superseded. Diff the version, trace each claim to its source, export, repeat.
Pricing
Start free. The staleness is the upgrade trigger: a free living review freezes 30 days after you stop paying — exactly when you'd most want it to keep moving.
One living review to feel the maintenance loop. Freezes 30 days after you stop paying.
For the researcher tracking the questions their work depends on.
For PIs defending published results across a wide front.
5-seat minimum. The shared review becomes your lab's canonical artifact.
Pricing shown is the planned launch tiering and may change before release. University site licenses are on the roadmap. Perennial is pre-launch — these are not live checkout pages.
Honest answers
Those are excellent tools, but they all search or look up. Elicit and Consensus run one-shot queries; Scite does claim-level entailment retrospectively, per paper. Perennial maintains a single living survey: it integrates each week's new work into the sections it affects and tells you which of your survey's existing claims are now supported, contradicted, or superseded. The job is maintenance, not search.
Before any paper is integrated, each reference it cites is resolved against Crossref and arXiv to confirm the DOI or arXiv ID actually exists. Sources whose references don't resolve — a hallmark of AI-generated and citation-mill papers — are filtered out. This is a deliberately concrete, checkable guarantee. Note that existence verification is different from, and more reliable than, automatically judging whether a paper supports or contradicts a claim; see the next question.
We treat them as carefully-flagged signals, not verdicts. Automated entailment on real papers is genuinely hard, so Perennial uses a two-stage design — a cheap first-pass classifier with a frontier-model adjudication only on flagged claim–paper pairs — and always links you straight to the quoted span so you can judge for yourself in seconds. You stay the expert; Perennial does the watching.
Perennial is strongest in preprint-native fields like machine learning and CS, where full text is openly available. Coverage in areas behind publisher paywalls is more limited because we only mine open-access and preprint full text. We'd rather be honest about where the product shines than over-promise across every discipline.
Perennial
Perennial is in development. Join the waitlist to be among the first researchers and labs to keep a literature review that never goes stale.