Errors, limits & scopes

Errors

Every error is:

{
  "error": { "code": "rate_limited", "message": "...", "details": {} },
  "request_id": "req_..."
}

Codes: invalid_request 400, unauthorized 401, forbidden 403, not_found 404, not_linked 409, rate_limited / usage_capped 429 (with Retry-After), upstream_error 502, internal_error 500.

not_linked means your platform identity has no account in that app yet - sign up there with the same email, then re-mint or refresh your key.

Pagination

List endpoints take ?limit= (1-100, default 50) and ?cursor=, returning { "data": [...], "next_cursor": "..." | null }. Cursors are opaque; pass them back verbatim.

Rate limits & AI caps

• 120 requests/min per key (standard), 20/min for AI-invoking endpoints.
• Daily AI caps per key during the beta: coach 25, cross-app bridges 10, pitch drafting 10.
• GET /v1/me shows your usage today and remaining limits.

Scopes

Scope	Grants
rndtoo:read / oos:read / regen:read	Read that app’s data (yours + reference)
*:write	Reserved; not granted during beta
ai:invoke	Coach, bridges, pitch drafting (capped)
platform:read	Cross-app aggregates (dashboard, year in review)
platform:admin	Operator tooling; implies everything

Beta keys get read scopes for every app linked to your email, plus ai:invoke and platform:read. Write scopes ship after the beta.

Health-content policy

Organic OS payloads and coach responses always include a disclaimer field. Interaction checks return severity framing with mechanism and recommendation, never a bare safe/unsafe verdict. Agents consuming these endpoints should relay the disclaimer.