LatteReview: configurable multi-agent literature screening

Python framework documenting customizable AI reviewers and multi-round literature-review workflows, including title/abstract screening and data abstraction. Reviewers can be arranged in parallel or sequential rounds using different model providers.

Workflow level n/a Source Repository-documented Outcome Not empirically evaluated here

Evidence & verification

Source level

Repository-documented
Grounded in a maintainer repository/README. Availability, stars and forks do not establish research benefit or correctness.

Outcome verification

Not empirically evaluated here
A documented resource or intended workflow, not a demonstrated academic outcome. No empirical benefit is established by its inclusion.

Sources checked

2026-10-17 (UTC)

Primary README-only source-semantic review. Capability assertions attributed to documentation; not-evaluated describes the adopted outcome claims, not an assertion the authors never evaluated their tool.

Source-checked does not mean independently reproduced. See the Method page for how these labels are assigned.

Sources (1)

Research applicability

Getting started

Editorial suggestion, not executed: read installation and the title/abstract example, inspect the licence for the selected version, then pilot a small labelled corpus and have researchers check exclusions and disagreements before scaling.

Maintenance

Live mutable HEAD README reviewed; no pinned release/commit or maintenance guarantee.

License

README says PolyForm Noncommercial 1.0.0 from 1.4.1 onward and CC BY-NC-ND 4.0 through 1.4.0; commercial permission needed. Licence file not independently audited.

Workflow & human oversight

Resource-level automation varies; no fixed workflow level assigned. Researchers define criteria, oversee data access and validate outputs. AI roles are not human editorial review.

Workflow level is not an evidence score; higher is not better evidence.

Approach

The README example passes title/abstract data through two initial AI reviewers and routes disagreements to a further AI reviewer using ReviewWorkflow. Additional documented reviewer types support scoring and abstraction; these are capabilities rather than independently verified outcomes.

Limitations

README-only review, not paper/full documentation or code execution. The README reports benchmark AUC and costs, but their datasets, calculations and generalizability were not adjudicated here and no numeric performance claim is adopted. AI reviewer roles are not human expertise. Researchers remain responsible for screening recall and source fidelity. Model availability and APIs may change.

Reproducibility

Source content saved and hashed; no install, execution or independent outcome experiment.

Adoption requirements

Python 3.10+, input articles and explicit inclusion/exclusion criteria; API credentials and budget for chosen hosted models, or a compatible local provider. Evaluate provider data policies before sending unpublished or restricted material. Commercial use requires licence review.

Comments

No comments yet.

Add a comment

Comments are held for moderation before appearing. Critical or corrective comments are welcome — especially evidence that an entry overstates what has been verified.