Skip to main content
Sentence-level evidence about brands. unit=mention: claims for and against, reluctance, with the sentence and context. unit=comparison: head-to-head comparisons with axis (price, audience, fit, standing, ease, deployment, integrations, support, performance, security, scope, features, other), the claim and winner (the brand that comes out ahead). sameAnswer requires other evidence in the same answer, e.g. praised for setup while a rival is praised for enterprise. Free: with credits every row; without, samples and aggregates (own brands in full).

Who may call it, and what it costs

Price. Free. Every reply says what the call cost in meta.costCredits. The prices are on Spending, tracking and stopping.

Arguments

one of mention, comparison
required
object
which answers count; window is a separate argument, not part of filter
list of string
brd_ ids; with brandRule, which answers count up to 20
one of named, recommended, first, not_named, warned
default named; warned: the answer advises against the brand by the caution-insight rule (warnsAgainst: the judge’s advised_against at least 0.65 and the brand not only a cited source; list_facets: mention.advised_against). It misses most warnings (recall 0.34) and is NOT the app’s Answers page rule, which reads the judged treatment warned_against (list_facets: mention.treatment)
one of any, all
with several brands; default any
list of string
up to 50
list of string
up to 50
list of string
up to 100
list of string
surface ids (list_facets: engine); without engine or surfaceClass, the counted engines up to 10
list of one of counted, wider, memory, search, imported
classes of engines to cover; default counted up to 5
"all" or list of string
plc_ ids (resolve_ids finds them): the place a question is about, with the places under each (a region holds its cities). “all”: every place, in place of a local subject’s area. A country code here reads as filter.country for one release
list of string
ISO country codes the question was asked from; an answer asked with no place set counts as US up to 20
list of string
domains or site: ids of cited sites up to 50
list of string
pg_ ids of cited pages up to 50
string
Full-text search over answer text: keywords, “exact phrases”, AND/OR/NOT, NEAR/n, prefix*. A call reads at most the first 100,000 matches (20,000 in search_answers) before the brand filter and the rest apply: a brand filter does not narrow it, so a common word over a long window can miss a brand’s answers; narrow by window, topic, market or engine
object
any key from list_facets
one of tracked, public, own, all
default all: public plus this workspace’s own; own: only the answers this workspace bought; tracked: its own answers and those about the questions, topics and brands it watches with the track tool (not the brands it added in the app); public: public answers only
object
default the last 28 days; “last” or from/to
one of 7d, 28d, 90d, 365d, all
date
inclusive
date
inclusive
one of for, against
list of one of none, conditional, fallback, budget_only, inferior_comparison, risk_warning, weak_praise
up to 7
list of one of price, audience, fit, standing, ease, deployment, integrations, support, performance, security, scope, features, other
up to 13
list of string
brd_ ids on the other side of a comparison up to 10
"self" or "other" or string
“self” (the filter brand), “other”, or a brd_ id
string
keywords over the sentence, its context and the claims (a comparison: its claim and what it is about): words, “phrases”, AND/OR/NOT, prefix*
object
list of object
up to 5
list of object
up to 5
one of company, product
how brands count: company (default; a company and its products are one brand) or product (each brand on its own, so a product’s numbers leave out its siblings)
integer
default 25; the token cap wins min 1, max 200
string
one of compact, full
compact (≈6k tokens, default) or full (≤20k)
list of string
columns to keep in each row up to 40
boolean
add one plain sentence saying what the reply shows
boolean
The schema as tools/list serves it, every shared object written out; https://agent.parse.gl/openapi.json serves the same schema with the filter and the window named once under components and pointed to by $ref:

Example call

The sentences in answers that argue against a brand. Over REST:
Over MCP, the assistant calls the tool by name with the same arguments:
The reply has data, context, next, page and meta; follow next for the calls that come after. A refused call returns { code, message, fix }: see the error table.