Litmus NoteKnow before you commit
About

Settings

Shown in the consent dialog and the disclosure statement.

Your own model key, stored in this browser only. Leave it empty to use the shared key configured for this deployment, if it has one. Needed for the model stages: polishing the statement, drafting queries, extracting fields, positioning, and drafting questions. Every stage also has a manual path.

Identifies your requests. Sent as the mailto parameter and recorded in each request URL.

Optional. A key raises the daily budget 10x, and searching works without one. Leave it empty to use the key this deployment already has (a Pages environment variable, or your local .env). Sent as an Authorization header, so it never appears in the recorded request URLs.

Optional here. Requests are throttled to one per second with exponential backoff. Leave it empty to use the key this deployment already has (a Pages environment variable, or your local .env).

Recorded verbatim in the AI-use disclosure. The tool makes no confidentiality promise of its own.

This stage sends your research idea and public paper metadata or abstracts to a third-party model. Nothing else leaves this browser.

Data minimisation: public metadata and abstracts only. Consent is timestamped and appears in the exported disclosure statement. You can cancel and keep working manually.

Method and limits

What this tool does

Litmus Note helps you decide whether a research idea is worth pursuing, by grounding it in an auditable literature search. The output is a Research Concept Note: evidence plus your own go / no-go decision.

The grounding rule

Every extracted claim carries a verbatim quote from the source abstract, checked against that abstract in this browser. A quote that does not appear in the source is rejected: the field is marked unverified and is excluded from the note. If the abstract is silent, the field reads "not reported in the abstract" instead of a plausible invention. This is enforced, not advisory.

Not

Standing disclosures