Skip to content

Insights

Insights, in the app’s sidebar for people with write on the project, and at GET /api/v1/insights, answers whether the brand is followed.

  • Brand answers per week: every file served, BrandHub listing read, use checked and search that found something, by people, agents and anonymous visitors.
  • Release adoption: fetches of the current version of an asset against fetches of a version that was already replaced when it went out.
  • Still on the old release: the replaced versions fetched lately, and the sites (referrer hosts) and agents that load them.
  • Most used assets, with their fetches by surface: the app, a key on the API, MCP, a portal, a signed link, a public embed.
  • Search gaps: searches with no result, from the app, portals and MCP.
  • Delivery and portal page views, per day and per page.

Its Use checks tab, at /insights/checks, is the use-check log: every check_use and POST /api/v1/check answer counted, refusals by reason (replaced, expired, territory, wrong variant), and the latest ones with who asked, what was offered instead and whether it was taken: the same client fetched the replacement, or checked it and was allowed, afterwards. Export CSV saves those rows.

Its Activity tab, at /insights/activity (from GET /api/v1/activity), is who did what in the project, newest first: uploads, reviews, deletes and brand changes, by people and by agents, each filterable.

Each asset’s panel has a Used in fold, from GET /api/v1/assets/{id}/insights: the brand rules that point at it, the brand pages that show it, the open public portals it is on, and its fetches over the last 30 days by surface and by the sites that load it.

Each brand has an Insights tab too, at /brands/{slug}/insights, from GET /api/v1/brands/{slug}/insights. A fetch or a check names a file, not a brand, so the brand’s files are the ones its rules held in its last 20 releases, each belonging to the newest release that holds it. From them: this week’s brand answers (its files served, uses of them checked, its BrandHub files read), the share asked by agents, the uses refused, and release adoption: the fetches since the latest release a day each, on it or on an older one (a file replaced in its stack counts as older), with the places that still load an older one this week, by site, agent or surface. The brand’s Overview shows the share on the latest release beside its pulls.

Connections, the page where agents are connected, adds what each one asked for over the last 30 days, from GET /api/v1/insights/connections, in a table: each agent with where it runs and its access (its key’s owner and scope), its calls, the MCP tool it asks for most (hover for every tool, how many failed, and the brand contexts it asked for), and what it was refused, by reason.

Two of the lists take a button, so a finding becomes something done. Fixes gathers both, with licenses about to end, in one list.

  • Notify, on each place in Where from, Old versions: an email to whoever answers for that site or agent, naming each replaced version it still loads and a /c/{id} link that always leads to the current version, so the next release reaches it without anyone asking again. Artbucket reaches the people it knows itself: a site or portal of the project goes to the project’s admins, an agent to the person who connected it. For anyone else (a site elsewhere, a key an admin made), name them in the dialog. Each version and place is told once: the row then says when (“Notified 2 days ago”), and a second notice for the same version is refused. POST /api/v1/insights/notices does the same, with referrer or client, the replaced assets and optional emails. It goes through the organization’s email and counts toward its daily emails, and the audit log keeps who sent it.
  • Request it, on each search in Asked for, Not found: an upload link, named after the search’s words, to send to whoever has the file, a teammate or an agency. What they send lands in Review, like any upload link’s files, and the gap says it was requested while the link is open.

Every install has it. A server that sells plans leaves fixes out of a plan’s features: Fixes, Notify and Request it then leave the app, and GET /api/v1/fixes and POST /api/v1/insights/notices answer 403. Insights itself stays.

Fixes, in the sidebar beside Review for the same people as Insights, is one list of what needs a hand, with the button for each, built from what Insights and the library already know:

  • A place still showing an old version: a site or an agent from Old versions, one row a place with every replaced version it loads, since the first day it did. Notify tells it, as above.
  • A search that found nothing, since the first day it did. Request it makes the upload link, as above.
  • A license about to end: an approved asset whose last day of use is in the next 30 days, soonest first. Open the asset to renew or replace it; once something replaces it, the row is done and Open the replacement leads there. Renewed or archived, it leaves the list.

What isn’t done comes first; what is stays below, saying when (“Notified 2 days ago”, “Requested just now”), while it is in Insights’ 30 days. The sidebar counts what isn’t done. GET /api/v1/fixes answers the same lists, with open, the count.

Every install has it. A server that sells plans leaves digest out of a plan’s features, and that organization’s projects are left out of everyone’s digest.

On Mondays, each editor and admin of a project gets an email about the week before, Monday to Sunday (UTC). It only has the sections with something in them, each linking where to act:

  • Brand releases: each brand released, its version and its release note.
  • New assets and new versions, counted, with the first few names.
  • Licenses ending in the next 14 days, soonest first, with what replaced the asset where something did.
  • Searches that found nothing last week, from Insights’ search gaps (only where Insights is on).
  • Waiting for review: how many assets, or suggestions, wait.

A week with nothing in any of them sends nothing. Each person gets a project’s digest once a week, whichever instance runs the sweep, and only what they could open in the app. Viewers and contributors get none.

It goes out with the organization’s email, so not at all where the organization can’t send (Settings, Email sending), and not from a read-only organization. Anyone turns it off in Settings, Notifications (PATCH /api/v1/me/digest), or from the Unsubscribe link at the bottom of the email, which needs no sign-in.

One row per event in the events table: a file fetched from /a/{id} (the app drawing its own thumbnails is left out), a check_use verdict with its reasons and the replacement offered, a search with words, a BrandHub file read, an MCP tool called (its name and whether it worked, never its arguments), and the brand context an agent asked for. Each keeps its kind, surface, whether the actor was a person, an agent or nobody in particular, an agent’s key name, the asset or subject, the version, the verdict, the referrer’s host, and the day.

Never an IP address, a person’s name or a full URL. There are no cookies or trackers for portal and hub visitors. Everything stays in your own Postgres: a self-hosted install sends nothing anywhere.

Raw events are kept 90 days. Once a day is over, it is rolled up into event_days, which is kept a year (the longest chart reads 12 weeks); charts read both through the event_counts view, so they are current whenever the rollup last ran.