Use the graph from your assistant

What it can look up

6 tools, listed here exactly as the server describes them to the assistant. The descriptions are long on purpose: they are the one set of instructions an assistant reliably reads before it answers.

Find a compound, outcome or proteinsearch_nodes

Search the Catalyst evidence graph by name or synonym (for example "magnesium glycinate", "sleep quality", "creatine"). Returns node ids to pass to get_findings or get_relations. It finds things; it does not say anything about them.

Graded findings about a nodeget_findings

The curated, graded study findings for one node: for a compound, what it has been shown to affect; for an outcome, which compounds have been studied against it. Each finding carries a grade from A (well established) to F (speculative), the verbatim sentence it was drawn from, the paper, the population and dose, and a permalink. A grade describes evidence strength, not a recommendation. Always give the grade with the claim, and link the permalink. An empty list means nothing has been curated yet, not that the compound does nothing.

One finding, with its grade explainedget_finding

One graded finding by its id (the last part of a permalink, /e/{id}): the claim, its grade and what that grade means, the verbatim quote, the study design and the paper. Use it to explain why a finding has the grade it has.

Reference-database relations (hypotheses)get_relations

What public reference databases (Reactome, UniProt, Rhea, ChEBI and others) record about a node: transporters, enzymes, pathways, expression. Every row is assertion_class 'inferred': a hypothesis about how a compound could act, not a result anyone measured in people. Present these as possible mechanisms and never as effects. For effects, use get_findings.

Reactions a compound or protein takes part in (inferred)get_reactions

Biochemical reactions from Rhea and Human-GEM that a compound is a substrate or product of, or that a protein catalyses or transports. Every row is assertion_class 'inferred': a reaction two public databases record, not a result anyone measured in a person. Present these as known biochemistry, never as an effect a compound has, and never as a recommendation. For effects, use get_findings. A compound flagged is_hub (water, ATP, protons and the like) returns its reaction count only, because it participates in nearly everything and a row list would not be biology anyone reads. Coverage: the whole of one Rhea release and one Human-GEM release, each reaction counted once; a participant that is not a node in this graph is listed but not linked, and only human enzymes that are nodes are listed at all. An empty result means no reaction in those releases names this node, not that the body has none.

What a compound can become, and through what (inferred, a hypothesis)reach

Breadth-first reachability through the reactions two public databases (Rhea, Human-GEM) record: what a compound could become, and through which reactions and enzymes, within up to 3 steps. Every route is assertion_class 'inferred': a hypothesis drawn from reference databases, never a measured or reported effect. Present it as known biochemical connectivity — never as an effect, a recommendation, or a claim that the body actually does this. For measured effects, use get_findings. Common cofactors and currency species (water, ATP, protons and the like) are excluded as intermediate steps, so a route never reads 'reaches everything through ATP'. A currency species is also never materialised as from_id or to_id itself — asking about one returns reachable: null with a coverage note explaining that, not a checked "0 routes". Give to_id to check one target compound; omit it to list everything from_id reaches. A route not being found can mean three different things, and the result's coverage field says which: reachability may not have been BUILT for this scope yet (nothing has been checked); it may have been checked and found NOT REACHABLE through the reactions loaded — never read that as "the body cannot make it"; or the compound asked about may be a currency species, structurally excluded rather than searched.

What every answer carries

1. 01

   A grade, and what it means

   Every finding comes with its grade, from A (well established) to F (speculative), computed from the study behind it and never set by hand. A grade says how strong the evidence is. It is not a recommendation to take anything, and nothing here is medical advice.

2. 02

   The sentence and the paper

   The verbatim sentence the finding was drawn from, the paper, the population and the dose, and a permalink to the finding on this site, so the person reading the answer can check it rather than trust it.

3. 03

   Findings and hypotheses, kept apart

   A connection assembled from reference databases is marked as a hypothesis in the data itself, and the assistant is told to call it one. It never arrives looking like a reported result.

4. 04

   Nothing for sale

   No tool returns a product, a price, a plan or a link to buy anything, and no tool ranks what to take. The only link an answer carries is back to the evidence.

How a grade is computed, with every ceiling and penalty, is on how it works.

Limits

Calls are counted per minute, per client and network address. ChatGPT and Claude each have 600 a minute, because one address there carries many people; editors have 120; any other client 60. A program that needs more can use a key (see access). The same graph is readable over plain HTTP at the API and as files from the export.

What is recorded

Not what you asked. The server does not store queries, and it has no account to attach them to. To enforce the limits above it counts calls per minute against the client’s name and a keyed hash of the network address, never the address itself. The rest is in the privacy notice.

Questions, or something an assistant got wrong from this graph: hello@catalystproject.ai.