ai_search
ChatGPTTranslate a natural-language query into a validated SciLeads filter and run the search. Consumes one AI-search credit. Use for the first search of a topic, or for conceptual refinements; prefer refine_search for structural changes (add a city, exclude an org). userQuery is capped at 500 characters - pass a focused search phrase, and summarise a longer ask rather than pasting it verbatim (an over-long query is rejected without consuming a credit). The response also returns two optional query improvements it did NOT apply — pluralVariants (grammatical forms of a term) and relatedTerms (broader/narrower concepts). Their exact JSON shape is in the output schema; offer them as follow-ups per the guidance below. Related terms and plural variants — offer as follow-ups, never up-front friction ai_search runs the search on the extracted filter term(s) only. It also returns two kinds of optional query improvements that were not applied. Present the results first, then — when it helps — offer these as follow-ups to broaden or narrow. Never gate results on them, never auto-apply them, and never comment on their quality (they exist to improve recall; the user judges relevance, not you). To actually apply one the user picks, add it via refine_search (structural) or a follow-up ai_search (conceptual). `pluralVariants` — offer FIRST (cheap recall). Plural/grammatical forms of a term (e.g. mutation → mutations) — the same concept, just pluralised, so the lowest-risk way to catch more results. The shape is nested so you can see each variant's source: the outer key is the extracted term, the inner key is the phrase the variant is a plural of — the base term itself, or one of its related terms. A variant off the base term is a pure recall win; a variant off a related term only matters once that related term is adopted. > "Want to include plural variants (e.g. mutation → mutations) to catch more results?" `relatedTerms` — offer AFTER plurals (broaden/narrow by intent). Ranked related concepts (broader or narrower), best first. On the SciLeads platform roughly the top 5 per term are pre-selected; here they are suggestions only, never applied. Frame them by what the user wants — focus vs widen — you need not use all of them, and you may add your own when they fit the natural-language query. > "I can narrow or broaden this — e.g. add pancreatic ductal adenocarcinoma to focus, or > carcinoma to widen. Want any of these?"
execute_search
ChatGPTRun a SciLeads search using a filter JSON (typically from ai_search or refine_search). Supports researchers and organisations. Returns up to pageSize results (max 50), total count, a deep-link URL, and a warnings array describing any dropped or Lucene-folded filter keys. Pagination is capped at 10 pages — beyond that, send the user to the OpenInPortalUrl for the full result set. Does NOT consume an AI-search credit.
find_filter_fields
ChatGPTSearch the curated catalog of SciLeads filterable fields. Returns matching field paths with usage descriptions, data types, and nested-query hints. Use to discover whether/how to filter on a concept (e.g. 'country', 'funding', 'researcher name'). Does not consume an AI-search credit.
generate_outreach
ChatGPTGenerate a tailored cold-outreach email message + subject for a single researcher using SciLeads' proven outreach prompt. Reuses the same generator (and per-user daily quota) as the SciLeads portal UI. Either set anchorOnRecentActivity: true to let the server pick the top 1-2 most-recent work items from the researcher's recent activity (recommended for "write me an email to X" flows after get_researcher_intel), OR supply explicit selected* ids (use when the user named a specific work item). The two modes are mutually exclusive; the combined explicit-ids cap is two. messageType currently only accepts "generic". When productId is supplied, the product name + description come from the caller's account catalog (same set get_researcher_intel returns) — the LLM cannot pass free-form product copy.
get_clinical_trial
ChatGPTFetch the detail envelope for a single clinical trial. Use the nctId returned by search_researcher_workitems for category=clinicaltrials, or any ClinicalTrials.gov NCT identifier. Default verbosity="summary" returns headline fields (titles, phase, status, dates, sponsor, conditions, interventions, and totals for every collection). verbosity="full" adds the brief summary, detailed description, eligibility criteria, study design, capped facility/outcome/mesh/keyword lists, and related-publication/grant rollups. Does NOT consume an AI-search credit.
get_current_user
ChatGPTReturn identity, role flags (account manager, demo mode), license expiry, and CRM authentication status for the calling user. Use to tailor suggestions — skip CRM workflows when no CRM is authenticated, flag expiring licenses, address the user by name.
get_funded_project
ChatGPTFetch the detail envelope for a single funded research project / grant. Use the awardId returned by search_researcher_workitems for category=funding, or any SciLeads award ID. Default verbosity="summary" returns the project facts (title, funder, dates, amount), the lead PI, and the organisation; verbosity="full" additionally returns the long-form description, public-health-relevance text, and keywords. Does NOT consume an AI-search credit.
get_publication
ChatGPTFetch the detail envelope for a single publication. Use the citationId returned by search_researcher_workitems for category=publications, or any SciLeads citation ID. Default verbosity="summary" returns headline fields (title, journal, identifiers, date, counts) and stays under ~4KB; pass verbosity="full" for the abstract, authors, mesh headings, keywords, topics, grants, and full-text URLs (each capped). Does NOT consume an AI-search credit.
get_researcher_intel
ChatGPTFetch the intel envelope for a single researcher: their facts, your account's product portfolio, the last 10 recent activity items, and recency triggers. Use as the data source for sales battle cards. Does NOT consume an AI-search credit. Returns { researcher, products, recentActivity, whyNow }.
get_tradeshow_poster
ChatGPTFetch the detail envelope for a single tradeshow poster or session. Use the posterId returned by search_researcher_workitems for category=tradeshows, or any SciLeads poster ID. Default verbosity="summary" returns the session, the presenting researcher, and the show; verbosity="full" additionally returns the poster abstract text and abstract-link URL. Honors the caller's HideResearcherEmail claim. Does NOT consume an AI-search credit.
open_in_portal
ChatGPTBuild a deep link the user can click to open SciLeads in the browser. Provide either (recordType + recordId) for a single record, or (entityType + filterJson) to open a saved-filter search.
refine_search
ChatGPTPatch an existing filter JSON without re-running natural language. Use for structural changes (add/remove a city, narrow a list); top-level patch keys REPLACE corresponding filter keys. Does NOT consume an AI-search credit.
search_organisation_workitems
ChatGPTSearch a single super-organisation's matched work items (publications, funding, tradeshows, clinical trials) for a given query. Use to drill into the per-org counts surfaced by execute_search for organisation entities. Each row carries the producing researcher's identity so the LLM can attribute findings or chain to get_researcher_intel. The response also includes top-level activity aggregates (distinct researchers, distinct trials, country count, total activity, email count) — these aggregates are scoped to the QUERY-MATCHED slice, not the org's full footprint. e.g. Pfizer's total clinical-trial count from execute_search will be higher than distinctClinicalTrials here once the query narrows the slice. Does NOT consume an AI-search credit.
search_researcher_workitems
ChatGPTSearch a single researcher's matched work items (publications, funding, tradeshows, clinical trials) for a given query. Use to drill into the MatchedPublications/MatchedFundedProjects/etc. counts surfaced by execute_search. Does NOT consume an AI-search credit.
