Summary

The Enrichment Agent turns your content into intelligent assets by sophisticatedly analyzing, tagging and summarizing assets then adds rich metadata that maps to your business’s unique taxonomy, opening up unlimited ways to find, reuse, and repurpose content across campaigns, markets and channels.

Who?

• Portal Administrators: Configure global portal instructions and manage agent access.

• Content Managers: select assets, run agents, and manage metadata suggestions.

Why?

Managing large volumes of digital assets requires consistent, rich metadata to keep content findable. Manual tagging at scale is time-consuming and prone to human error. The Enrichment Agent evaluates visual and contextual content, maps it directly to your established portal taxonomy, and suggests metadata unique to your business. The result is a scalable way to maintain a clean, searchable asset library.

• Saves time: eliminates manual data entry for titles, descriptions, and tags, accelerating asset onboarding.
• Improves content discoverability and reuse: extracts visual elements, text, logos, and attributes, and enriches assets with business-aware metadata, making assets more discoverable, reusable and ready to activate
• Maintains consistency: maps attributes directly to your predefined taxonomy options.
• Enables personalization of content: Deliver personalized content experiences dynamically using asset meta properties based on targeting and segmentation.

What type of enrichment workflows can it facilitate? 

• Core asset: Naming creation and conventions for asset name, file name, descriptions, alt text
• Multilingual: Translations
• Creative and brand: Visual style, color palette, background and setting, shot type and framing, time of the day
• Content classifications: content category, product attributes, animal breed attributes, market/region, location, language, campaign, collection 
• Technical: File format, aspect ratio, resolution, color space 

How?

Creating an Enrichment Agent

1. Open the Agent Library via Settings > AI Agents > Manage AI Agents.
2. Click Create agent in the top-right corner.
3. Select agent type Enrichment.
4. In the agent editor, enter a natural language prompt describing the exact task you want the agent to fulfill. See the guide How to Write a Good Prompt.
5. Use the toolbar for additional features:
   • Insert metaproperty: specify metaproperties the agent should read or update, including where the result should be saved (the asset's Title, Description, or a specific metaproperty). Type / while writing your prompt to bring up the metaproperty selector.
   • Add knowledge: attach up to 19 assets that give the agent extra context that would be impossible, or just too inconvenient, to include in your prompt. Instead of asking the agent to guess based on general knowledge, you ask it to apply your specific business knowledge. See the guide How to Best Use Attachments.
   • Improve prompt: one-click optimization in the editor. The agent adds structure, specificity, and best practices automatically. Review the improved prompt before saving; you stay in control of the final instructions.
6. (Auto)generate a Name and Description, then click Create Agent to save it to your library.

For agent statuses (Draft, Available, Disabled) and auto-apply, see Agent Library: Advanced features.

Running the Agent

1. Go to your Asset Bank and select the target assets you want to enrich.
2. Select the Enrichment Agent you want to run.
3. Start the run. Processing happens in the background and you'll be notified when it's done.

Reviewing Suggestions

1. Navigate to the Agent Activity page to view a complete log of agent output.
2. Suggestions that require review appear under To-do. Suggestions that were auto-applied or previously accepted appear under History.
3. To undo an entire agent run, go to the Agent History view and click Undo. This reverses all accepted suggestions at once and restores assets to their previous values.

Tip: If the output looks off, don't delete it. Keep the results and contact us; this helps us investigate and improve.

Limitations and Technical Constraints

We work closely with customers to identify bottlenecks and use that input to guide improvements. Several items below are already on our roadmap. Let us know if a limitation is blocking your workflow.

Selection Limit

• Maximum Assets per Run: you can apply an agent to up to 10,000 assets in a single run.

Supported asset types and sizes

Metadata handling and default behavior

• Agents can suggest values for metaproperties, asset title, and description.
• By default, the following metadata is always shared with the agent as context, whether or not your prompt mentions it: the asset's title, description, tags, and the metaproperty option (MPO) values the asset is already tagged with.
• Agents can also read additional metadata fields such as original file name, date added, and archive status. Mention these fields in your instructions if you want the agent to use them.
• When you use the / selector in a prompt, all available MPO values for that metaproperty are shared with the agent to guide its choice.

For metaproperties (MPOs):

• All metaproperty types in Bynder are supported.
• Only the first 1,000 metaproperty options (alphabetically, A to Z) are read by the agent.
• Agents overwrite existing metaproperty values with new suggestions by default. Appending, re-using, or merging with existing values is supported if the agent is instructed to do so.
• Agents cannot create a new MPO when no matching option exists. This limitation does not apply to Text metaproperties.
• Use Metaproperty definitions to give more context to the agent. Click here to learn more about how to do that

For tags:

• Tags have their own syntax in your prompt; use / or insert them using the toolbar button.
• Agents overwrite existing tag values by default, but can keep, append, or merge existing values if instructed to do so.

Change tracking

• A log of all agent changes is available on the Agent Activity History page.

Language support

Our AI agents support a broad range of languages, with the strongest performance in widely spoken languages and meaningful capabilities maintained across languages with fewer digital resources.

Capabilities:

• Multilingual asset processing: agents can analyze and generate metadata in multiple languages.
• Automatic language detection: no configuration required.
• Multi-language assets: agents can process assets containing text in several languages.
• Flexible prompts: Global AI Instructions and agent prompts can be written in any supported language.

Controlling output language:

1. Write your prompt in the target language (for example, "Analysiere das Bild und erstelle einen Titel").
2. Explicitly specify the language in the prompt (for example, "Analyze this image and generate metadata in French").
3. Set it organization-wide in Global AI Instructions (for example, "Always generate metadata in Spanish unless otherwise specified").

Performance considerations: AI models are trained on digital sources, so languages with larger digital footprints (such as English) show the strongest accuracy and cultural nuance. Less common languages still provide meaningful capabilities, but you may notice reduced accuracy in complex or nuanced tasks.

Related Articles

How to Write a Good Prompt

How to Best Use Attachments (Add Knowledge)

Workspace: Agent Activity - Reviewing, Editing, and Approving Agent Suggestions

Updated July 27, 2026 12:43