Skip to main content
Fourteen tools. Your assistant picks between them; this page is for when you want to know what it actually has access to, and why a call was refused. Every tool is scoped to the organizations you ticked at consent and runs with your own role in each. A tool your role does not permit is refused, exactly as the dashboard would refuse it.

​
All fourteen

​
Reports and queries

run_report runs one of roughly fifty named reports — revenue, conversion, AOV, attribution, funnels, top products, collection performance, returns, orders by channel or gateway or province, customer cohorts, data integrity. These are the same reports the dashboard shows, computed by the same code, so a figure in chat and a figure on the page cannot disagree. query composes instead: any measure by any dimension, filtered and bucketed. Reach for it when the question is a notch off a named report — “revenue by country by week for the last 90 days, mobile only”.
Which reports a store can answer depends on which Glood apps it has connected. A store with Glood Search but not Glood Personalization answers a different set. That is what describe_reports and describe_query are for, and why asking what exists beats guessing at a name.

​
Tools that look alike

They answer different halves of a catalogue question.explore_catalog returns Glood’s behavioural metrics — recommendation rank, click-through, collection-to-product-page movement. “Which products convert best” is this one.shopify_graphql returns Shopify’s own records — inventory levels, fulfilment status, metafields, order detail. “Which of them are out of stock” is this one.A question like “which of my best-converting products are out of stock” needs both, in that order.
run_report runs one of the fifty named reports. Prefer it when one fits, because those are what the dashboard shows.query composes a breakdown no report covers. It has a larger vocabulary and more ways to answer a subtly different question than the one asked, which is why the glood-query skill exists.Comparisons belong to the tool either way: both take a comparison window and compute it. A trend should never be two calls subtracted.
agents is what an agent is — its instructions, schedules, deployments and versions. Read this before proposing any change.activity is what it did — runs, the findings it raised and how often each recurred, what it read during a run, and what it currently believes. “What did it find last night” is this one.manage_agent changes things. Every write reads back a sentence naming the agent, the store and the organization.

​
Shopify access is read-only

shopify_graphql runs against your own Shopify Admin API, and mutations are refused — checked on the parsed document, not guessed at from the text. There is no way to write to Shopify through this connection. Customer identifiers are refused by name: email, phone, names and addresses are not returned. If something needs changing in Shopify, the assistant will tell you what and where. You make the change in Shopify admin.

​
Writes that ask twice

Three actions are two-step by design. The first call changes nothing and returns a review of what would happen; nothing happens until you confirm:
  • Archiving an agent
  • Stopping a deployment
  • Removing a schedule
Everything else that changes an agent reads back a sentence naming exactly what changed, on the store and organization it changed it on. That sentence is your confirmation the right target was hit — read it.

​
Creating an agent is not deploying one

Creating makes the agent and its instructions. It then runs nowhere and recurs never until you deploy it to a store and give it a schedule. Cron runs in each store’s own timezone.