Calibrated AI auto-excludes the clear nos in your systematic-review search — with an audit trail librarians can defend. Your team only reads the uncertain middle.
Incumbents sell a collaboration workspace. syrev sells workload removed — with proof.
Rayyan · Covidence · DistillerSR
Beta free · then per-review / seats
ASReview etc.
Scan left to right — green column is syrev.
| Capability | Rayyan / Covidence style | syrev.ai | ASReview |
|---|---|---|---|
| Collaboration UI | ✓ | beta | local |
| AI ranks / suggests | ✓ | ✓ | ✓ |
| Auto-exclude clear nos | ✕ | ✓ | ✕ |
| Auditable decision log | ~ | ✓ | ~ |
| Replay your review as proof | ✕ | ✓ | ✕ |
| Entry price (public lists) | Free → ~$60–100/seat/yr or $339+/review/yr |
Beta: 1–2 free then $149 / review |
$0 (self-host) |
Competitor prices from public pages (Sep 2026) — verify before publish. Dual-reviewer is on our roadmap, not in first beta.
Clear nos leave the queue. Reviewers spend time where judgment matters.
Per-criterion probabilities and a threshold you can put in the methods paragraph.
Jev for speed and calibration; DeepSeek proven on the same public sets with matching recall.
Default threshold exclude_below = 0.005. Recall = fraction of truly included studies not auto-excluded. DeepSeek matches Jev on recall; Jev is faster.
~75% auto-excluded · recall 1.0 · 1,443 abstracts. Cite the paper; we don’t claim an open CSV dump.
68–70% auto-excluded · recall 1.0 (Jev / DeepSeek) · 2,359 abstracts. CC BY 4.0.
~12% auto-excluded · recall 1.0. Low savings when criteria are hard to see in abstracts — we show that case on purpose.
We’ll show auto-exclude rate, missed includes if any, and hours saved — then walk the uncertain pile with you. Every replay includes a methods paragraph and audit appendix you can paste into a protocol.