crawl_find platform and kind: which Sume social route each calls

Sume's agent tools group the 15 social REST routes into four calls: crawl_profile, crawl_feed, crawl_media and crawl_find. Each takes platform (instagram or tiktok), and crawl_feed and crawl_find also require kind, which picks the underlying route.

The names do not always match the REST route. In crawl_find, platform: tiktok with kind: videos runs keyword search, not a video lookup, and kind: profiles runs tiktok_search_users.

The mapping

crawl_profile and crawl_media need only platform. crawl_feed and crawl_find need platform and kind. An unsupported pair, such as instagram with top, is rejected with 'Unsupported platform/kind for this crawl job.' before any lookup runs.

Input is validated against the same schema as the REST route, so every REST field (handle, limit, max_pages, order) carries over unchanged.

Why it matters when you write a prompt

An agent that is told to 'find TikTok videos about a topic' should call crawl_find with kind: videos, and one told to 'list this creator's videos' should call crawl_feed with kind: videos. Same word, different tool. Spell out the tool and kind in your Format instructions if the distinction matters to the output.

Slow reads come back as a job; the agent continues with jobs_wait and then fetches the result, as described in the OpenAPI text for these routes.

What these tools do not do

They read public data only: no login, stories, DMs or private accounts, at most 24 items and 3 pages per call, and nothing is imported on its own. To keep a video, the agent calls media-imports_create; to understand what a clip contains, it uses video_inspect, both listed in the tools and gates page.