Features

Watching

The part no metadata tool does.

Frame sampling: 36 stills spread evenly across each video, not the thumbnail. Production quality, setting, on screen text, and any product held up to camera are read from the frames themselves.
Transcript reading: The full spoken transcript, so what a creator claims can be checked against what the frames show. When the two disagree, that gap is reported rather than smoothed over.
Audience response: The top comments by engagement. Real questions and shared experiences read differently from emoji noise, and that difference is scored.
Three platforms: YouTube with full frame coverage, TikTok and Instagram through cover frame, caption and counts. The engine is told when frames are thin so it never describes visuals it did not see.

Scored against your brand, not a generic quality bar.

Brand context: Your pillars, audience, voice, content guidelines and off limits topics. Every judgement the agent makes is made against these, including how the outreach reads.
Contextual brand safety: Context, not keyword blocklists. A dermatologist explaining a prescription drug is safe. A creator discussing their own recovery responsibly is safe. Each score carries the reason.
Creator content profile: Five dimensions synthesised across a creator's videos: visual style, communication patterns, topic expertise, authenticity markers, safety indicators.
Discovery intelligence: Production trajectory, audience authenticity and expertise depth. Surfaces creators whose craft has outrun their follower count instead of ranking by reach.
Evidence on every score: Each judgement links to the specific video it came from. You can check the reasoning rather than trusting a number.

From shortlist to a message you can actually send.

Outreach drafting: A note that cites a specific moment from a specific video, matched to that creator's register. Capped at 120 words, no marketing adjectives, no mail merge tells.
Contact discovery: Business emails extracted from the descriptions and bios already collected. Nothing is guessed: a returned address is one a human published, or the field is honestly blank.
Gmail handoff: One click opens Gmail with the recipient, subject and body filled in. Virlia has no send path at all, so nothing leaves without you pressing send.
Parallel analysis: The whole shortlist is processed concurrently against identical criteria, so a run takes about as long as its slowest creator rather than the sum of all of them.

Once the ranking exists, the work is narrowing it.

Bring your own list: Paste up to 500 channels or videos you already have and they are judged against your brief the same way discovered creators are. The parse is shown before anything is spent, because analysing the wrong 500 is the one mistake that cannot be undone.
Look-alikes: Given one creator who fits, find others who work the same way. It ranks candidates worth a proper run rather than returning a verdict, and says so.
Head to head: Two creators from a shortlist compared directly on the same criteria, for the decision that comes down to a choice between two rather than a ranking of thirty.
Lists and tags: Named sets of creators that persist across runs, so a shortlist you spent an hour on is not something you rebuild next campaign.
Blocklist: Creators a brand never wants to see again, filtered out of every future run rather than rejected repeatedly.

What happens after the message goes out.

Pipeline stages: Prospect, contacted, replied, negotiating, contracted, live, delivered. Held separately from the fit judgement, because a creator you passed on can also be one you paid last quarter.
Reply logging: Paste what a creator sent back and it is classified and filed against them: what kind of reply it is, the sentiment, what they asked for, and any rate they quoted, with the currency and what it is per.
Deliverables and submissions: Briefs a creator can open from a link and submit a draft against, with no account and nothing to install. Reviews and comments happen in your workspace.
Content library: Collected from what creators actually submitted rather than uploaded by hand, and searchable afterwards.
Decisions that carry forward: Every approve, reject and star is recorded against the brand, not the run, so the signal accumulates across campaigns instead of resetting each time.

Getting the work out of the tool.

Shareable reports: A read-only link to a run, for the people who need to see the shortlist and the evidence without an account.
Exports: The shortlist, the scores and the contacts, out in a form a spreadsheet understands.
White label: Your own name, logo and accent on a shared report. On Agency, because it exists for people showing work to their own clients.