Every claim traces to a paper you can open, or the agent says it found nothing.
Tamats searches PubMed for peer-reviewed studies and the open web — general, scholar and news — and returns results with the title, source and link attached, never a claim with nothing behind it. A search that finds nothing says so and stops rather than filling the gap with a recalled fact, and any paper worth keeping saves to the organisation's shared library with its citation intact. PubMed is the tool for evidence; general and news web search cover trend and regulatory-change questions PubMed was never built to answer, and the agent is instructed to reach for the right one.
Most cosmetic research questions have a real answer sitting in a paper, not in a model's memory of one. Finding it and keeping the citation attached is the tedious part — a formulator asking what the evidence says about an ingredient needs an answer with a source, not a fluent paragraph that might be right. Tamats treats literature search as retrieval, not recall: every result is a real hit against PubMed or the web, with somewhere for it to land once you've found it.
What you do with it
PubMed search, scoped to cosmetics and dermatology
Plain keyword queries are automatically scoped to the cosmetic/derm corpus and ranked for relevance; a search that returns nothing can be retried once with the scope dropped before the agent abstains. Results carry title, authors, journal, year, DOI and a link.
General, scholar and news web search
Separate from PubMed: general search for broad lookups, scholar for papers PubMed doesn't index, news for recent regulatory or market changes, and shopping/image/video modes for competitor references. Every result is cited with a title, link and snippet.
Save to a shared library
A paper worth keeping saves to the organisation's library in one action — no re-typing the citation. The library starts empty and holds only what the team actually saved, split by paper, formula, image and finding.
Findings keep their lineage
A retrieved paper stays attached to the query and the claim it supported, so a finding in a later conversation can still be traced back to the search that produced it.
What it computes over
—Live PubMed queries via the corpus package, not a cached or memorised result
—Live web search results (general/scholar/news/shopping/images/videos) with source links
—The organisation's own saved-paper records in its library
Where it stops
PubMed and web search only — there is no full-text access to paywalled journals; a search returns the abstract and a link, not the paper itself.
A search with no PubMed hits abstains rather than answering from the model's own memory of the literature — that abstention is a finding, not a failure to fix.
Saving a paper stores the citation and metadata your team chose to keep, not the corpus itself — Tamats does not host or redistribute journal content.
Who picks it up from here
The researcher who reads the saved paper and decides whether it actually supports the claim being made — search finds the source; it doesn't interpret it.
Questions
Does literature search ever answer from the model's own memory instead of a real paper?
No — a query that returns nothing from PubMed is reported as nothing found, retried once with broader scope, and then abstained on. It never falls back to a recalled fact dressed up as a citation.
What happens to a paper after I save it?
It lands in the organisation's shared library with its citation intact — title, authors, journal, year — searchable alongside every formula, image and finding the team has saved, with no re-typing required.