Embed a custom view
Once you’ve created and published a custom view, you can embed it in your own app: Paid renders the dashboard in a sandboxed iframe, and a signed token scopes the data to exactly one of your customers. Only per-customer custom views can be embedded this way; organization-level custom views are for internal use in Paid and are never embeddable, regardless of their status. The customer never logs into Paid, and the underlying queries are filtered to that customer in the database, not in the browser.
This is the recommended, easiest integration: your backend asks Paid for a short-lived token using your API key, and the @paid-ai/embed SDK handles the rest. You never manage a signing secret.
Custom views are experimental. Routes, token claims, and SDK surface may change without notice and are not covered by the v2 backwards-compatibility guarantees.
Prerequisites
- A published, per-customer custom view (an organization-level view can never be embedded) and its
displayId(e.g.view_5oi3qvfaFLK). See create a custom view with Claude. - A Paid API key, used server-side only.
Step 1: Fetch an embed token from your backend
Your backend calls the Paid embed-token endpoint using your API key and returns the token to your frontend. Never expose your API key to the browser.
The endpoint mints a short-lived signed token scoped to a specific customer:
Response:
customerId accepts the customer’s Paid display id (cus_...), external id, or internal UUID.
Calling this endpoint for an organization-level view’s displayId returns 400 with code ORG_SCOPED_VIEW_NOT_EMBEDDABLE, since organization-level views are never embeddable.
Option A: use the SDK (recommended)
If Paid publishes an SDK for your stack, use it. The typed method makes the request for you and returns the token and its expiry. Call it from your own authenticated backend route and return token / expiresAt to the frontend.
TypeScript
Python
Go
Ruby
Java
Option B: call the REST endpoint directly
If there’s no SDK for your stack, call the endpoint over HTTP from your backend:
Node.js
Python
Step 2: Embed with the SDK
Install the embed library:
Pass a getToken callback that fetches a fresh token from your backend. The SDK calls getToken for the initial token and again automatically whenever a token expires. You do not need to manage refresh yourself.
renderCustomView returns a handle:
token is the view’s displayId (which dashboard to show); getToken
returns the signed access token (which customer to scope to). They are
different values, do not pass the JWT as token.
Pass a date range from your app
By default the embed renders the view’s stored date range (the timeframe set in Paid). If your app has its own date picker, drive the embed’s range from it: post a paid-embed:period message to the iframe whenever your picker changes, and the dashboard re-renders in place with the new window.
The period payload accepts the same two shapes as the view’s stored timeframe:
Send period: null to return to the view’s stored default range.
If you build the embed URL yourself instead of using the SDK, set the range with query parameters on the JWT-backed embed page (/public/views/{displayId} opened with a ?token= embed token): periodKind=relative&periodUnit=day&periodAmount=7, or periodKind=absolute&periodStart=2026-07-01&periodEnd=2026-07-31. Share links generated from the Paid dashboard (/public/custom-views/...) do not accept these parameters.
The range only affects views whose SQL uses the {period_start} /
{period_end} placeholders (views created with a timeframe do). The range is
a display setting, not an access boundary: the data always stays scoped to
the customer in the signed token, whatever range is requested.
Pass filters from your app
A view can declare named filter parameters when it is created (for example a region filter with the values it accepts). If your view declares filters, your app can drive them the same two ways as the date range.
Declare the filter on the view and reference it in the SQL:
An unprovided filter binds the empty string, so the = '' branch is the match-all default.
Then push values from your page whenever your own controls change:
Send filters: null to clear back to unfiltered. On a direct embed URL, pass the same values as query parameters instead: filter_region=eu.
The embed’s view endpoint returns the declared filters alongside the render bundle (each filter’s name and, when set, its accepted values), so your app can build its filter controls from the declaration instead of hardcoding them.
Only declared filters are accepted: an unknown name or a value outside the
declared list returns 400 with the reason. Like the date range, filters
narrow data within the token’s customer: they never widen access, and they
are not an isolation mechanism.
Discover a customer’s hierarchy
For an organization with customer hierarchies enabled, your backend can read the active units available to a published per-customer view. The API key needs read:analytics; an authoring-only key cannot read unit names.
customerId accepts the customer’s Paid display ID, external ID or internal UUID. Each directory entry contains externalId, name, externalType, parentExternalId, isRoot, billable and effectiveBillable. billable is the unit’s own setting; effectiveBillable is false when the unit or any unit above it is non-billable. Both describe the unit now. A change applies only to signals received afterwards, so classify past usage by each signal’s own billing mark (data._billing.excluded), not by these fields.
Use the response’s hasHierarchy field to distinguish customers with active child units from customers with only a root or deleted children. A child with no usage still qualifies. This value describes the whole customer, even when the current page contains only the root or no entries.
Follow pagination.hasMore, increasing offset by limit, to load the remaining entries. The maximum page size is 100. Build parent relationships from the complete directory; a parent can be on another page. Deleted units are excluded, and names and relationships reflect the next request after a rename or move.
The directory is also available from /api/public/views/{displayId}/customer-units with a signed embed token in the Authorization: Bearer header, and /api/public/custom-view-links/{token}/customer-units for a live shared link. These routes derive the customer from the credential and accept only pagination parameters. They use the same publication, expiry and revocation rules as the view.
Hierarchy discovery returns structure only. It does not add a selector to the dashboard or filter the view’s datasets. When customer hierarchies are disabled for the organization, discovery returns 403 FEATURE_NOT_ENABLED.
Label units inside the view
The view itself does not need to call the directory. For a hierarchy-enabled, per-customer view, every paid:data message the dashboard receives carries a context object (version 2) with the customer’s units and the current selection. This applies equally to the Paid preview, an embed and a shared link.
Select the unit key in your query, for example ifNull(external_customer_unit_id, '') AS unit, and look up its name when the message arrives:
unitSelectionis{ "mode": "all" }, or the selected unit’sexternalIdandnamewithdirectorsubtree.unitslists only the viewing customer’s active units, at most 1,000, oldest first. Above that,unitsTruncatedistrueandunitsTotalis the full count; fall back to the key for any unit not listed.- A deleted unit is not listed, but its history still counts in its ancestors’ subtree totals, so keep the key fallback.
- Views without hierarchy navigation, and organizations without customer hierarchies, receive no
units. Older bundles that read onlydata,unitSelection.mode,unitSelection.externalIdandperiodkeep working unchanged.
Show cap usage per unit
A view can also receive the customer’s unit caps as caps in the same context. This is opt-in: create or update the view with capsEnabled: true, alongside hierarchyEnabled on a per-customer view. Turn it on only for views that draw cap usage, since each load then reads cap usage. Views without it receive no caps. Caps are spend policy, so the API key that reads the view’s data or mints its embed token also needs read:billing: data read without it leaves caps out, and minting returns 403 INSUFFICIENT_SCOPE. A token records this when it’s minted, so only a token from a key with read:billing shows caps; any other token, including one minted before caps existed, never does, even if they’re turned on later.
Each entry is one cap in force on one of the customer’s units and credits currencies. Its fields are exactly what the customer unit cap endpoints return, and used covers the unit and every unit under it.
Work out the share used and whether a unit is over its cap in the view, so you choose the rounding and how exactly 100% reads:
- A unit can have one cap per credits currency, so match caps by
externalIdandcreditsCurrencyIdtogether. - At most 25 caps are sent, in the directory’s order. Above that,
capsTruncatedistrueandcapsTotalis the full count. For more caps, or caps that follow the unit selection, querydim_customer_unit_cap_currentinstead (see spend caps). currentPeriodisnullwhen no usage was reported for that cap.- When cap usage can’t be read in time, the view still loads:
capsis left out andcapsUnavailableistrue. Show the rest of the view and leave the cap figures out rather than showing zeros.
If the view stops serving
An embed only resolves for a view that’s currently published. If the view is still a draft, or a previously published view is later unpublished, the embed endpoints return 404 (NOT_FOUND), exactly as they would for a view that never existed, and your customer sees nothing render.
An already-issued embed token isn’t automatically invalidated by unpublishing, but that no longer matters: any request it makes still resolves against the view’s current status, so the embed stops rendering data as soon as the view is unpublished, without waiting for the token to expire.
Unpublishing is available from the view’s page in the Paid dashboard and takes effect immediately. It doesn’t delete the view, so republishing it (from the dashboard, from Claude, or via the API) makes the embed start resolving again with no changes needed on your end.
How scoping and isolation work
- The customer comes only from the verified token’s
subclaim, never from a client parameter. An unknownsub, a missing or invalid token, or a token signed for the wrong resource is rejected. - The view’s queries run against a per-customer database layer that filters every row to the token’s customer. The author’s SQL contains no customer filter; isolation is enforced in the database.
- The dashboard renders in a sandboxed iframe with no network access of its own. The signed token is never passed into that frame, only the already-scoped data for that one customer.