Up to 50 (platform, handle) pairs → one canonical Author per row, failed handles refunded.
Batch profile lookup for handle vetting. POST a JSON body with an items array of 1-50 { platform, handle, custom_id? } objects (handle accepts an @handle, a bare handle, or a pasted profile URL); returns one row per item: the canonical Author object (same shape as the single GET profile endpoint), a status of ok / not_found / unsupported / error / deferred, the caller's custom_id echoed, and the credits that row cost: in input order. Per-item isolation: one dead or private handle never fails the batch. Billing is per item at each platform's own credit tier (most platforms 1 credit; LinkedIn 5); only successful rows are charged, everything else is refunded. Supported platforms: instagram, tiktok, youtube, twitter, threads, twitch, snapchat, truthsocial, bluesky, kwai, reddit, telegram, linkedin (a member, or a company page as a linkedin.com/company/… URL), facebook. A LinkedIn lookup that is rate limited is retried once and then answered deferred at no charge: re-request it. Never cached. Optionally streams Server-Sent Events when the client sends Accept: text/event-stream. Send include: "posts" (up to 25 items) to add each handle's first page of posts beside its profile: every ok row gains a posts block with the same post rows, next_cursor and has_more as that handle's own post list endpoint (endpoint names it: instagram profile/posts, tiktok profile/videos, youtube channel/videos, twitter user/tweets, threads user/posts, facebook profile/posts, reddit profile/posts, telegram profile/posts, linkedin profile/posts or company/posts; other platforms answer posts.status: "unsupported"). since on the body and stop_at_id on an item stop each page at a date or at a post you already hold, as on those endpoints, and posts.stopped_at says which. Add posts_limit (1 to 100) to walk each handle's post list until that many posts are held: posts.pages is the pages charged and posts.stopped says why the walk ended (limit, end, window, page_cap). include: "posts,about" also returns activity (last_post_at, posts_30d, posts_90d, media_mix) and stats (avg_likes, avg_comments, engagement_rate) on every row, computed from the posts held. Add comments to include (it implies posts) to read the comments of each handle's top posts: the comments_top_posts posts (default 3) with the most comments among the posts walked (a post without a comment count is ranked by views after them, or rank all by views with comments_select_by: "views") each gain comments with items (the same comment rows as that platform's comment endpoint: instagram, tiktok and facebook post/comments, youtube video/comments, twitter tweet/replies, linkedin post/comments, reddit post/comments, threads post/comments), total (the post's comment count), truncated and next_cursor, up to comments_per_post (default 20) per post; the row's comments block gives status (ok, partial, failed, deferred, empty, unsupported) and the credits. One post that fails never drops the row. The profile in data is unchanged.
Request
- Base URL: https://www.socialcrawl.dev/v1
- URL: https://www.socialcrawl.dev/v1/prism/profiles
- Auth: API key in header x-api-key
Headers
Send no-cache to bypass the response cache and force a live fetch. Billed at the normal endpoint cost; the fresh result is written back to cache for the next caller. Only the no-cache directive triggers this. See the Response Schema guide for details.
Optional UUID that makes the request safely retriable. The key is reserved atomically per account before billing. A concurrent retry receives 409 IDEMPOTENCY_IN_PROGRESS with a Retry-After of 1 second while the original is young, growing to a quarter of the time it has been running (at most 30); a crashed reservation expires after 5 minutes. A replay keeps the cached payload immutable except for billing metadata: credits_used becomes 0, idempotent_replay becomes true, and credits_remaining is refreshed to the current balance. A known current balance appears in both the body and X-Credits-Remaining header; no balance row resolves to 0. On a transient lookup failure, body credits_remaining is null and X-Credits-Remaining is omitted. Completed outcomes expire after 24 hours. If the claim store is unavailable, the request fails closed before billing. GET endpoints do not combine idempotent replay with Server-Sent Events: use Accept: application/json when sending this header. The unsupported key-plus-SSE combination is rejected before billing.
Request body
Response
Successful response
Example response
{
"success": true,
"platform": "prism",
"endpoint": "/v1/prism/profiles",
"credits_used": 1,
"credits_remaining": 4999,
"request_id": "req-abc123",
"idempotent_replay": true,
"meta": {
"hint": {
"code": "search_multi",
"message": "You searched \"matcha\" on 3 platforms one call at a time; /v1/search/multi returns each platform's own search rows in one call at the same price per platform, and a platform with no rows is not charged.",
"method": "GET",
"path": "/v1/search/multi?query=matcha&platforms=instagram,tiktok,youtube"
}
}
}Changes
No recorded changes to this operation across all 1 revision of this API.