AllNutrition

Get the research evidence for a nutrition question from AllNutrition's vetted library, then write the answer from it yourself. Use this for any nutrition, diet, or supplement question where the user needs a trustworthy, citable answer rather than a guess from the open internet. Sources are trust-scored, and industry-funded or biased studies are filtered out. Returns an evidence pack (plain text), not a finished answer: - source-quality labels: evidence_strength (strong | moderate | limited | insufficient) and consensus_level (high | moderate | mixed | low) of the retrieved sources; - answer guidelines: how to weigh study designs, the audience, and the citation format; - numbered sources [1], [2], …: title, authors, journal or publisher, year, URL, evidence level, design, trust score, a one-line methodology note, and excerpts. Write the answer from the pack following its guidelines: cite sources as [n], end with a References list of the sources you cited with their URLs exactly as given, and state the evidence_strength and consensus_level labels. Excerpts are published abstracts or openly licensed (CC BY) full text only. The pack contains no personal data and nothing about the caller's AllNutrition account. The question and the sources are saved to the caller's own private Ask history in the AllNutrition web app by default (visible only to their account, never shared); saving can be disabled per call (``save_session=false``) or account-wide in the web app under Settings → API keys. Args: question: The nutrition question, in plain English (max 1000 characters). Make it self-contained: include the population, condition or dose the user mentioned earlier in the conversation. published_after: Optional ISO-8601 cutoff (e.g. ``2020-01-01T00:00:00Z``) to only use evidence published on or after that date. deep_research: Return a larger pack for a comprehensive, sectioned research report instead of a short answer. Use it when the user asks for deep research, a comprehensive report, or an in-depth review, and pass ``search_queries`` with it. search_queries: With deep_research=true, 2-5 targeted searches you write yourself, each covering a distinct sub-topic of the question in its own keywords (for omega-3 and heart health: "omega-3 supplementation cardiovascular events trials", "fish oil atrial fibrillation risk", "EPA DHA dose triglycerides"). Ignored for a standard ask. save_session: Set false to skip saving this question to the caller's private Ask history. Omit to use the default (save, unless the account has turned saving off in its settings).

ask_nutrition

Search AllNutrition's curated reference library — ranked sources, no LLM answer. Peer-reviewed sources ranked by semantic relevance × composite trust score, deduplicated to one entry per source. Use this when you want raw citations to feed into your own pipeline or to surface evidence directly in a UI. Returns exactly these fields, and no others: query echo of the search query count number of references returned results[] id, title, url, publisher, publisher_short_name, evidence_level, trust_score, published_at, passage, source_basis, trust_basis ``passage`` is a short AI-generated summary — verbatim source text is never returned. ``source_basis`` / ``trust_basis`` (``full_text`` | ``abstract``) say whether the match and its trust score rest on the full text or only the abstract. No field contains personal data or anything about the caller's AllNutrition account. Args: query: Free-text search query. max_results: Maximum references to return (1–20, default 10). published_after: Optional ISO-8601 cutoff to only return newer sources.

search_references