Actions from chat — marking, running analyses, seeing inside a post
Beyond reading, Claude can take three kinds of action through Huntit.
The same protections hold on all three. Anything that costs money says so up front, and a whole class of these calls costs nothing at all. Analyses can't double-run. Nothing reachable from chat can delete a client, a profile or a post.
Marking posts
Discussing a post and want to keep it? Ask Claude to mark it favorite or reviewed — the same triage marks your team uses in the app, shared with everyone. Free, instant, reversible.
Claude can read them back too, so "show me everything we starred but never briefed" is a real question now.
One honest limit: Huntit stores the mark, not the date you set it. So "what did we mark this month?" is answered over the posts' own publish dates instead.
Running an analysis
If a post's intelligence brief, comments intelligence or custom analysis hasn't been generated yet, Claude tells you and offers to run it.
Approve, and the run behaves exactly like clicking the button in the app: it saves to Huntit, so your whole team sees the result, and it uses your agency's own Gemini credits at the same cents-level cost.
Claude tells you a run is billable before it starts. The per-run estimate lives in the app, on the comment-depth panel, not in the chat.
On a video over the 5-minute limit it refuses the brief and the custom analysis outright — nothing started, nothing billed — and offers comments intelligence instead.
Timing is honest. Huntit holds the conversation briefly, then hands Claude a free follow-up read that waits again — so a Brief still arrives in the same conversation, and it saves into Huntit either way.
Comments intelligence takes a few minutes, because the comments get collected first, so Claude starts it and checks back. A run already in progress can't be started twice.
Looking inside a post — video and carousels
Two different paths, and they cost differently.
A video. Claude can't watch video itself, so Huntit lends it eyes.
Ask "look at this video and tell me how the transitions work" and Huntit has the video watched by Gemini, which sends back a rich visual description: scenes, pacing, on-screen text, motion. Claude then reasons over that description, focused on whatever you asked.
This uses your Gemini credits, typically a few cents, and Claude tells you it's billable before it runs.
A photo carousel. A different and better path: Claude can already see still images, so Huntit hands it the slides — all of them, in order — and Claude reads them itself. No AI ever runs on a carousel.
On TikTok the whole read is free. An Instagram or Facebook link from outside your tracking still spends one scraping credit, because fetching the post is how Huntit finds out it was a carousel.
What's kept, and what's capped
A watch is never kept: the description lives in your conversation and nowhere else, never entering your Huntit data. If a watch outruns the wait, Huntit holds the finished description just long enough to hand it to Claude, then drops it.
The 5-minute cap is one limit across Huntit, not a watch-only one. Over it the AI never watches the media: no automatic read when the post is found, and no brief or custom analysis either.
A video whose length Huntit can't read is refused the same way, before anything is billed. A carousel has no length to cap — every slide always comes back, never a first few.
Tracked posts and outside links
Both work on a post you already track and on a link from outside your tracking:
- A tracked post — for a carousel, the slides ride the post's normal detail read: free, instant, no AI.
- An outside link — a TikTok, Instagram, Facebook, or YouTube post from a profile you don't track. Anything that might be a video is confirmed first, because a watch costs credits. A TikTok photo link copied from the browser is the exception: it is answered straight away, free, with nothing to confirm. A TikTok share link — the short
vm.tiktok.comkind — is confirmed first, then answers that nothing was billed after all. An Instagram or Facebook link is confirmed either way, and costs one scraping credit whichever it turns out to be.
If you find yourself looking at the same outside profile repeatedly, that's the signal to add it to tracking — then everything about it is collected, analyzed, and kept.
One quirk worth knowing: on claude.ai the slide images are occasionally not readable in the same turn that asked for them. Ask again in the next message and they're there.
The safety rails, plainly
Huntit's guarantees don't depend on the AI: analyses can't double-run or double-charge, and there is no destructive action at all — nothing reachable from chat can delete a client, a profile, or a post.
There's no daily limit, deliberately: it's your own Gemini key, so the pace is yours to set.
What stands in its place is visibility. Huntit keeps a running count of the paid runs made from chat in the last 24 hours, and Claude states that count back to you as it climbs. The one hard stop is a guard against a runaway loop, set far above anything real use reaches.
The approval step you see before each action is Claude's own current behavior, on top of those.