Learn how to choose and configure Hugo’s AI model settings based on answer quality, cost, compliance, and reliability.

Hugo can use different AI model families and processing options depending on your workspace needs. These settings help you balance answer quality, speed, credit usage, hosting region, fallback behavior, and media processing.

Model availability can evolve over time. Always use AI Agent → Agent → Settings as the source of truth for your workspace.

In this guide, you will learn how to:

• Understand Hugo’s model settings → see what each option controls
• Choose automatic or manual model selection → decide how Hugo should pick its model
• Compare available models → understand model families, intelligence, price, and hosting
• Choose a hosting region → align Hugo with your privacy and compliance needs
• Enable the fallback model → keep Hugo reliable when a provider is unavailable
• Allow image and audio processing → let Hugo understand screenshots and voice messages
• Test your setup → validate your configuration with real customer questions

Understand Hugo’s model settings

Hugo’s model settings define how your AI Agent processes customer conversations. They control which model Hugo uses, where requests can be processed, and how Hugo should behave when extra input such as images or audio messages is sent by users.

You can find these settings in Crisp under AI Agent → Agent → Settings.

Main settings available

From this section, you can configure:

• Model → choose automatically or select a specific model
• Region → choose the processing region when available
• Built-in fallback model → let Hugo fall back to a Crisp-hosted model if needed
• Image processing → allow Hugo to analyze images sent by users
• Audio transcription → allow Hugo to transcribe audio messages before answering
• Response style → choose how Hugo should structure its replies
• Answer guidance → decide how cautious Hugo should be when answering
• Agent locale → let Hugo auto-detect the customer language or use a fixed locale
• Data sources → define which sources Hugo can use when generating answers

Choose automatic or manual model selection

You can either let Hugo choose the model automatically or manually select a specific model.

Choose automatically

By default, Hugo can choose the model automatically. This is the recommended setup for most teams because it gives Hugo a strong default model without requiring you to compare every option manually.

The current auto-selected model is GPT-5.1 from OpenAI.

This is a good starting point if you want strong answer quality and do not have specific internal requirements around model provider, hosting region, or cost control.

Select a specific model

You can also choose a specific model manually.

This is useful when your team has specific needs around:

• Compliance → using a model or region approved by your internal policy
• Cost control → choosing a lower-cost model for high-volume support
• Provider preference → selecting OpenAI, Anthropic, Google, Mistral, or Crisp-hosted models
• Performance testing → comparing answer quality across models
• Language coverage → testing which model works best for your customer languages

Compare available models

Before choosing a specific model, use the comparison table available in Crisp.

Go to AI Agent → Agent → Settings, then click How to select the best model? below the model selector.

This table helps you compare the available models based on:

• Model provider → OpenAI, Anthropic, Google, Mistral, or Hosted by Crisp
• Intelligence → expected reasoning and answer quality level
• Price → relative credit usage
• Hosting → available processing regions

Model families available in Hugo

If you want to understand how model usage affects your credits, read this guide: How does Hugo AI pricing and billing work?

Choose a hosting region

The Region setting lets you choose where model requests should be processed when region options are available for the selected model.

For supported models, you can choose between:

• Default (Global) → standard processing region
• European Union → EU-based processing when available

This is especially important for teams with stricter privacy, security, or compliance requirements.

EU processing

When European Union is available for a model, Hugo can process requests in the EU for that selected configuration.

This can help teams that need stronger control over where AI requests are processed, especially for corporate, regulated, or privacy-sensitive environments.

Data retention and compliance expectations

For supported configurations, Crisp is designed to process AI requests with strict data-handling requirements, including configurations where provider-side data retention is not used.

If your company has specific compliance requirements, confirm the selected model, region, and data processing expectations with your internal legal or security team before going live.

You can also read: Crisp EU GDPR compliance status

Enable Hugo’s built-in fallback model

Hugo can use a built-in fallback model hosted by Crisp.

This fallback model is used when the selected model cannot process a request, for example during a provider outage, temporary limitation, or unavailable model response.

Why fallback matters

The fallback model helps keep Hugo available even when the selected model provider cannot respond.

It is mainly a reliability layer.

Enable the fallback model if you want to:

• Reduce interruptions → avoid blocking Hugo when a provider has an issue
• Improve reliability → keep conversations moving when the selected model is unavailable
• Protect support continuity → make sure customers still receive an answer whenever possible
• Add a safety net → use Crisp-hosted fallback processing when needed

If Hugo behaves unexpectedly, you can also check this guide: Troubleshooting & Common questions about Hugo AI Agent

Allow image and audio processing

Hugo can process more than text messages. You can allow it to analyze images and transcribe audio messages when customers send them.

Allow Hugo to process images

When image processing is enabled, Hugo can analyze images sent by users and use them as context before answering.

This is useful when customers send:

• Screenshots → error messages, settings, or product issues
• Interface captures → where they are blocked in your product
• Visual proof → billing, delivery, or account-related screenshots
• Product images → if visual context helps answer the request

Allow Hugo to transcribe audio messages

When audio transcription is enabled, Hugo can convert audio messages into text before processing them.

This is useful for support channels where users often send voice messages, especially on mobile or messaging channels.

Common use cases include:

• WhatsApp support → users sending voice notes
• Mobile users → customers who prefer speaking instead of typing
• Accessibility → users who cannot easily write long messages

Configure response style and answer guidance

Model settings also include options that influence how Hugo answers.

Response style

Response style controls how Hugo formats and structures replies. For example, Hugo can answer in a more conversational style or keep answers more direct.

Answer guidance

Answer guidance controls how cautious Hugo should be when answering.

A conservative setup helps Hugo answer only when it can rely on available sources and when the answer is likely to be correct.

These settings are useful, but they are not the main way to control Hugo’s behavior. For detailed behavior rules, tone, escalation, and answer structure, use instructions.

Set the agent locale

The Agent locale setting controls the language behavior of Hugo.

In most cases, Auto-detect is the best option because Hugo can adapt to the customer’s language automatically.

Use auto-detect

Use Auto-detect when your customers contact you in multiple languages or when you want Hugo to respond in the same language as the customer.

Use a fixed locale

Use a fixed locale if your support experience should always stay in one language, or if a specific team handles a specific market.

Recommended setup by use case

There is no single best configuration for every workspace. The right setup depends on your support volume, customer language, compliance needs, and quality expectations.

Test your model settings

Before using a model configuration in production, test it with real customer questions.

Go to AI Agent → Evaluate → Playground and compare how Hugo answers with your current setup.

What to test

Use examples from real support conversations, including simple, complex, and unclear questions.

When testing, check:

• Answer accuracy → does Hugo answer correctly?
• Source usage → does Hugo rely on the right training data?
• Reasoning quality → does Hugo handle complex or ambiguous requests well?
• Escalation behavior → does Hugo involve a human when needed?
• Speed → is the answer fast enough for your support experience?
• Credit usage → does the model fit your expected volume?

If answers are weak, do not only change the model. Review your training sources and instructions first.

Useful guides:

• How to train Hugo AI Agent on your data
• How to prompt Hugo AI Agent for Instructions

Frequently Asked Questions

Still have questions which were not covered in this article? Here is a collection of the most frequently asked questions on this topic.

Which model does Hugo choose automatically?

The current auto-selected model is GPT-5.1 from OpenAI. This may evolve over time, so always check AI Agent → Agent → Settings for the latest value.

What does Hugo’s built-in fallback model do?

The fallback model helps Hugo continue answering when the selected model cannot process a request, for example during an outage, limitation, or provider issue.

Should I enable image processing?

Enable image processing if customers often send screenshots, interface captures, or visual context that Hugo can use to answer more accurately.

Should I enable audio transcription?

Enable audio transcription if customers send voice messages, especially on mobile or messaging channels such as WhatsApp.

Is changing the model enough to improve Hugo’s answers?

Not always. Model choice matters, but answer quality also depends on your training sources, instructions, routing, and test process. If Hugo gives weak answers, review your knowledge and instructions before switching models.

Updated on: 28/05/2026