InverityNine agents that never stop working
The agents are a standing system, not an import step. They work while assets are being created, again every time one changes, and continuously across everything already in the library. Each can run on its own, propose changes for review, or stay switched off.
A library that only does what it is told stops being useful at scale. Somebody has to tag the asset, notice the duplicate, spot the missing release, and chase the reviewer who never answered. That work is unglamorous and endless, and it is the first thing dropped when the team is busy. It is also exactly why asset libraries decay into shared drives with better branding.
The agents do that work continuously. They are not an import step that runs once and hands back a tidy library which immediately starts rotting. They evaluate work while it is being created, re-evaluate it on every version, and keep sweeping assets nobody has opened in months. How much autonomy each one gets is your decision, not a default you inherit.
The agents
Enrichment Agent
Adds tags, alt text, descriptions, and color palettes, then keeps them accurate as an asset is edited and re-versioned rather than only when it first arrives.
Brand Compliance Agent
Checks assets against your guidelines and flags what falls outside them.
Librarian Agent
Keeps the library tidy: duplicates, naming, missing metadata, misfiled assets.
Governance Agent
Watches license windows and flags assets no longer cleared for use.
Transformation Agent
Derives the formats, crops, and renditions each destination needs.
Design Review Agent
Scores creative work against your criteria and leaves structured feedback.
AI Search
Answers plain-language questions about the library and returns the assets.
Agent Configuration
Set what each agent does, when it runs, and what needs a human to confirm.
Asset Approval Agent
Moves assets through your approval route and chases outstanding decisions.
How it works
When the agents act
While the work is still being made
Agents do not wait for a finished file. The Design Review Agent scores work in progress against your criteria, and the Brand Compliance Agent flags drift while there is still time to change it cheaply. Feedback arrives during creation rather than after delivery.
Every time an asset changes
A new version is not a blind spot. Enrichment re-runs, compliance re-scores, and the record updates itself, so metadata and approval state describe the asset as it is now rather than as it was when it first landed.
Continuously, across the whole library
The Librarian Agent keeps working on assets nobody has touched in months, catching duplicates, naming drift, and metadata gaps. The Governance Agent watches license windows and talent releases and flags assets whose rights have lapsed, whether or not anyone is looking at them.
At the moment someone asks
AI Search answers plain-language questions against the current state of the library, and rights and approval status travel with the results. What comes back reflects what is true now, not what was true at import.
Whenever a decision is outstanding
The Asset Approval Agent routes work to the people who own the decision and keeps chasing the ones that have not answered, so approvals do not stall quietly in an inbox.
On whatever terms you set
Each agent can run on its own, propose changes for a human to confirm, or stay switched off. Agent Configuration is where you decide which of those applies, per agent and per workspace.
Questions about the agents
Do agents act without approval?
That is your choice, and it is set per agent. Each one can run on its own, propose changes for a human to confirm, or stay switched off entirely. Starting with everything in propose mode and promoting individual agents once you trust their output is a reasonable way in.
Can we adjust what an agent looks for?
Yes. Agent behavior is configurable per workspace, including the rules the compliance and governance agents evaluate against. The guidelines you publish in the Brand Portal are what the compliance agent scores from, so tuning the guidelines tunes the agent.
What happens when an agent gets something wrong?
Agent output is a suggestion or a flag attached to the asset rather than a silent change, so a wrong call is visible instead of buried. Version history means anything an agent did can be rolled back to what preceded it.
Do they run on our existing library, or only on new assets?
Both. Because they are continuous rather than triggered only by arrival, assets that have been sitting untouched are evaluated too. Switching on a new agent applies it to what you already have, not just to what arrives next.
Does this replace our reviewers?
No. It removes the mechanical part of review: checking a palette, spotting a missing release, catching a duplicate. Your reviewers still make the judgement calls, but they make them with the mechanical checks already done and the context attached to the file.
How are our assets handled during enrichment?
Handling of your assets, including any subprocessors involved in enrichment, is covered on our security page.
Book a demo and see the agents run against your own assets.