Policy-driven by default
Every action an agent takes is subject to the same brand, licensing, and content rules a human reviewer would apply — enforced as the work happens rather than at a final gate.
Layer 1 — agents inside the DAM
Cloudinary's DAM agents
This is what agentic DAM looks like when it stops being a category description and becomes a product. Cloudinary ships four specialist agents — Taxonomy, Search, Moderation, and Workflow — fronted by a Coordinator Agent that reads your intent and deploys whichever of them the job needs.
They were announced in May 2026 and are built on the same Model Context Protocol servers Cloudinary brought to the DAM market in June 2025 — which is why the second layer of this story, covered on Claimable Clouds & Agent Experience, runs on exactly the same foundation.
Taxonomy Agent Search Agent Moderation Agent Workflow Agent Coordinator Agent
One door in
You talk to the Coordinator, not to four tools
The single entry point. Describe the outcome you want and the Coordinator interprets the intent, then deploys the right agent — or the right sequence of agents — to deliver it.
- 1
State the outcome
A sentence in plain language — no query syntax, no knowledge of which agent owns which step.
- 2
The Coordinator plans
It interprets the intent, decides which specialists are involved, and sequences them.
- 3
Specialists execute
Taxonomy structures, Search retrieves, Moderation judges, Workflow moves the result downstream.
- 4
You review the result
A finished deliverable plus the record of how it was produced — and a flag on anything that needs a human.
Structure
Taxonomy Agent
Builds and maintains the metadata structure — keeping tags consistent, values normalized, and the taxonomy safe to evolve as the library grows.
Asked in plain language
“Add custom fields for expiration and model name to this collection.”
- Proposes and applies structured metadata fields against real business logic
- Normalizes tag values so the same concept is never recorded three ways
- Extends an existing taxonomy without orphaning the assets already filed under it
- Produces the structured metadata that downstream AI discovery depends on
Discovery
Search Agent
Finds assets by intent rather than by field, so nobody has to know the schema to retrieve the right approved image.
Asked in plain language
“Show me a final, approved product shot of a handbag from our summer season that can be used for a print campaign only.”
- Translates a plain-language request into a governed, permission-scoped query
- Reads usage rights, approval state, and licensing windows as part of the match
- Returns assets that are safe to use for the stated channel, not just visually similar
- Works across languages without a separate index per market
Governance
Moderation Agent
Evaluates assets for quality, consistency, and compliance against brand guidelines and content rules — approving, rejecting, or escalating with its reasoning attached.
Asked in plain language
“Reject anything with a stock watermark; flag low-quality shots for review.”
- Checks incoming and user-generated content against brand standards in real time
- Returns an explicit decision — approve, reject, needs review — plus the reason for it
- Routes genuine edge cases to a human reviewer instead of guessing
- Leaves an audit trail behind every automated verdict
Orchestration
Workflow Agent
Turns a plain-language description of a process into a governed automation spanning ingest, transformation, approval, and delivery.
Asked in plain language
“When an asset is uploaded to the Products folder, generate alt text, add it to metadata, and send the info to Shopify.”
- Builds multi-step automations from a sentence, not a flowchart editor
- Chains conditions, AI analysis, metadata writes, and downstream delivery
- Connects Cloudinary to the CMS, PIM, commerce, and martech tools already in place
- Keeps every generated workflow inside existing permissions and policy
What makes the autonomy safe
Agency is only useful if it is bounded
The reason these agents can be trusted with production libraries is not that they are cautious — it is that every action is policy-driven, permission-aware, and configurable, with human review reserved for the cases that genuinely need judgment.
Permission-aware, not privileged
Agents inherit the scoping of the person or system they act for. A search cannot surface an asset the requester was never entitled to see.
Humans on the edge cases
Ambiguous decisions route to a reviewer with the agent’s reasoning attached, so a person judges the hard 1% instead of processing the easy 99%.
Auditable end to end
Automated decisions carry a trail: what was changed, by which agent, on what basis. Autonomy stays reviewable instead of becoming a black box.
Why the boring layer matters
Agents are only as good as the metadata underneath them
Every capability on this page rests on structured metadata: seasons, SKUs, usage rights, approval state, expiry dates. That is why the Taxonomy Agent exists and why it comes first in the lifecycle.
When those fields are consistent, a request like “approved, in-window, print-only” becomes a query an agent can actually satisfy. When they are not, an agent guesses — confidently and wrongly. The unglamorous work of normalizing a taxonomy is what converts a media library into something autonomous systems can reason about, which is also what makes a brand portal like this one trustworthy for the people browsing it.
How enrichment on ingest works
The other half of the story
These agents live inside the DAM. The DAM also has to work as a tool for agents outside it.
Layer one is Cloudinary's own agents acting on your library. Layer two is the reverse direction: a coding assistant, a chat interface, or an agent inside your own product provisioning a Cloudinary environment and calling media operations as typed tools. Same platform, opposite direction of control — and you need both before “agentic DAM” means anything durable.
Sources
- Cloudinary Agents: AI That Acts Across Your Visual Media Lifecycle — Cloudinary
- Cloudinary Launches AI Agents to Streamline Enterprise-Scale Visual Media Management and Brand Governance (May 5, 2026) — Business Wire
- Cloudinary AI Agent Tools and MCP Servers — documentation — Cloudinary documentation
Agent names, descriptions, and dates on this page follow Cloudinary's own published material. Screenshots are Cloudinary product imagery. For current capabilities and availability, see Cloudinary's announcement.