Errors & limits

The request did not include valid credentials.

The credentials are valid but lack permission for this operation.

Email/password combination did not match any account.

The provided API key did not match a known prefix, or its hash did not validate.

The API key has been disabled by an administrator.

The authenticated account is not the owner of the requested resource.

The request originated from a domain that is not in the account's allowlist.

The end user has not granted the consent required for this operation.

The authenticated account is missing required setup data.

Too many requests. Check the `Retry-After` header for the recommended wait time. The `extensions.scope` field hints at the partition (`user`, `ip`, `session`, `api_key_slot`).

The account already has the maximum number of active sessions.

The waiting room queue is at capacity.

The account has consumed its allotted quota (minutes, knowledge units, storage, etc.).

This operation requires an active paid subscription.

Sessions cannot start at the current time per the persona's schedule.

Returned only as WebSocket close 4005, never as HTTP. The live session reached the persona's per-session interaction cap (PublishSettings.SessionInteractionLimit, counted in billing units — Full mode = 2 per turn, Chat mode = 1).

Returned only as WebSocket close 4006, never as HTTP. The live session exceeded the persona's per-session duration cap (PublishSettings.SessionDurationLimitSeconds, wall-clock — includes Standby/idle time).

The uploaded file exceeds the per-endpoint size limit.

The file content does not match the declared content type (e.g. an `.exe` renamed to `.pdf`).

The file's content type is not accepted by this endpoint.

The document exceeds the maximum page count for ingestion.

No resource exists at the addressed URL, or it is not visible to the caller.

The operation conflicts with the resource's current state (duplicate, version skew, etc.).

A turn is already being processed for this session. /v1/chat serializes writers, so the second caller is rejected before anything persists.

A live (WebSocket) surface currently owns this session, so it cannot be driven over REST at the same time.

The brand still has sources, so deleting it removes every document of every source. The delete is refused until the brand's display name is sent back as `?confirm=`.

One or more of the brand's sources are an integration's streams. An integration is an account-level connection, so a brand delete does not sever it.

An onboarding job is still building a persona and crawling knowledge under this brand.

Only one bulk knowledge job runs per account at a time, and one is already queued or running.

No source of this name exists in the account. A write never creates a source implicitly.

A source of this name already exists under a different brand of the account. Names are unique per account.

The source is one of an integration's streams: only that integration writes it, so it is read-only to the dashboard and the account key. That is what keeps it 1:1 with the connected site.

A source of this name already exists under the same brand but holds the other thing — a catalog where files were asked for, or the reverse. A source's type is fixed when it is created, so this is a different source under a taken name.

A source holds a catalog or files and never both. Typed entries go into a catalog; an uploaded file, page or post goes into a source of files. The route and the source disagree.

A document carries at most `Knowledge:Acceptor:MaxLibraryEntries` images and as many links (15). The list sent was longer; nothing was dropped and nothing was written.

The source still has documents, so deleting it removes them all with their category schemas. The delete is refused until the source's name is sent back as `?confirm=`.

An onboarding's crawl is queued or running into this source. A running crawl has no cancellation hook, and a queued one is minutes from writing.

One onboarding at a time per account. The response carries the running job's `jobId`, `brandId` and `personaId`.

The scrape of the site's homepage failed — the site refused, timed out, or returned nothing usable.

The source named by the site's host already exists as a catalog, or belongs to another brand of the account.

Only one bulk knowledge job runs per account at a time, and one is already queued or running.

A build of this source's summary is already queued or running.

Another write created this record at the same moment, so this one was not saved.

The source is being imported right now.

The source's last import skipped nothing, or its site sends what was skipped itself (a WordPress stream).

The integration was disconnected on personaizer.com, so its streams are frozen read-only and its key no longer writes.

The site pushed into a door that is shut: the integration is disconnected, or the stream is switched off or was never switched on.

The source name this stream would use already belongs to another source of the account.

The stream's full list names a generation older than the one last applied, so applying it would undo newer syncs.

The widget persona chosen on the consent screen speaks for a different brand than the one approved.

One or more request fields failed validation. The `errors` extension contains per-field details.

An unexpected server-side error occurred.