Overview

AI Configurations in Akeneo PIM are designed for performance, reliability, and scalability. Because AI generation involves analyzing structured product data, assets, and prompts simultaneously, a few technical and practical limits apply.

These limits help ensure a consistent experience for all customers while maintaining data integrity and predictable performance across large catalogs.

This page explains:

• The maximum supported configuration sizes
• File type and asset restrictions
• Attribute option limits
• General performance and reliability considerations
• BYOK-specific considerations (model behavior, fallback, costs)

Configuration limits

The following limits apply to AI Configurations in your Akeneo PIM environment.

Recommendation: If you manage multiple teams or product lines, organize configurations by purpose (for example, "Enrichment," "Translation," "Audit") rather than creating one configuration per attribute. This keeps the total number manageable.

Attribute option limits

AI Configurations can generate values for select-type Attributes such as Simple Select or Multi Select. These Attributes have a maximum option limit.

Best practice: If you have a large list of select options, consider grouping or filtering them in advance to keep the set below 500.

Supported asset types

When assets (images or documents) are included in an AI Configuration, Akeneo PIM automatically analyzes them to extract contextual data — for example, product color, certifications, or logos. Only certain file types are supported.

Asset quantity and size limits

Each product can include a limited number of assets when running an AI Configuration, to maintain performance consistency.

Asset collection requirements

AI Configurations only process assets stored in collections that include a main media Attribute. This ensures that each asset has a defined primary image or document for analysis.

If you include a collection without a main media reference, it is skipped during execution.

Processing order for assets

When multiple assets are available, the AI processes them in the following order:

1. Images from the selected Asset collection (up to 5, in collection order)
2. PDF documents from the selected Asset collection (up to 2, in collection order)

This order helps ensure the AI uses the most relevant visuals first — for example, product packshots or hero images — before supplemental documentation.

Reference Entity context limitations

When a Reference Entity Single Link or Multi Link Attribute is used as an input source in an AI Configuration, Akeneo PIM automatically includes the linked record's attributes in the prompt payload, scoped to the selected Channel and Locale.

The following attribute types within the Reference Entity record are supported:

• Text
• Text Area
• Simple Select
• Multi Select
• Number

The following attribute types within the Reference Entity record are not sent to the AI:

• Images and Assets
• Reference Entity Single Link / Multi Link

Fair usage and performance management

Akeneo monitors AI feature usage to ensure equitable access and performance for all customers. There are currently no hard usage limits, but usage is tracked internally to prevent abuse and manage costs.

If your team consistently runs large-scale automation, contact your Akeneo representative to evaluate suitable scaling or quota options.

When using the BYOK add-on, Akeneo's fair-usage limits do not apply — but your AI provider's own rate limits and quotas take effect instead.

Recommendations:

• Run enrichment rules in batches rather than simultaneously.
• Schedule heavy automation during off-peak hours.
• Use Preview and manual runs to test configurations before scaling.

BYOK-specific considerations

The following considerations apply when using the Bring Your Own Key (BYOK) add-on. If you are using the default Akeneo-managed model, you can skip this section.

No automatic fallback

If your selected BYOK model cannot process a request (invalid API key, expired credits, provider outage), the generation or translation will fail. Akeneo does not automatically fall back to the default model.

To minimize disruption:

• Keep your API keys active and monitor credit balances with your provider.
• If a critical Configuration needs high reliability, consider keeping a backup Configuration set to the default model.

No built-in cost dashboards

Akeneo does not currently provide cost tracking, usage dashboards, or spending caps for BYOK models. Monitor usage through your provider's tools:

• OpenAI: platform.openai.com/usage
• Anthropic: console.anthropic.com

Model behavior differences

Different models may produce noticeably different outputs for the same prompt:

Temperature settings and thinking models

Anthropic Claude models (Claude Opus 4.6, Claude Sonnet 4.6) use a "thinking" approach to generation. For these models:

• Custom temperature settings in your Configuration are not applied. The provider's defaults are used instead.
• This is handled automatically — no action is required from you.
• In practice, Claude models may produce more consistent outputs compared to OpenAI models at higher temperature settings.

Provider rate limits

Each AI provider enforces its own rate limits (requests per minute, tokens per minute). Large-scale Rules Engine automations with a BYOK model may trigger rate-limiting errors.

• Start with smaller rule scopes and increase gradually.
• Check your provider account's rate-limit tier and upgrade if needed.
• If rate-limit errors occur, affected products will not be enriched in that run.

Choosing the right model for your use case

Data privacy and security

All AI enrichment runs through Akeneo's enterprise AI infrastructure. Your data is never used to train AI models, never shared with third parties, and never stored after processing. All data is transmitted over encrypted channels.

When using the BYOK add-on, data is routed to the provider you selected (OpenAI or Anthropic) using your own API key. Your key is encrypted at rest and transmitted only at execution time.

For full details on how your data is protected — including provider-specific policies and API key handling — see AI Configuration Security & Data Privacy.

Key considerations for reliable performance

Troubleshooting common limit issues

Summary

Understanding AI Configuration limits in Akeneo PIM helps you:

• Design efficient configurations
• Avoid data overload
• Maintain high enrichment performance
• Scale confidently across your entire catalog
• Choose the right BYOK model for each use case

For detailed recommendations on AI output behavior and realistic expectations, see Understanding AI Behavior and Realistic Use Cases.