Model selection and management

What’s included in this article

• Model selection overview
• How to add external models 
• Managing models
• Best practices
• FAQs

Model selection overview

With model selection you can choose the right AI models for defined agent use cases across your org, while managing them through a unified governance layer — ensuring confidence in model choice, routing, and compliance enforcement. For technical configuration details, see our dev docs.

How to add external models

From AI Studio navigate to Models & Guardrails > Models > +Add model to get started. 

Once you’ve launched the modal for adding a model you will be taken through various steps where you will need to add credentials from your model provider. Please take a look at our dev docs for more details here.

Once you’ve added your model provider information, you can select who has access to use this model when building agents. You can grant access to All teams which will give anyone with builder access the ability to leverage that model when building agents. To restrict access to specific groups, select Specific teams and then add the team(s) that should have access to this model.

Managing Models

Once you’ve added a model you’ll see it appear in the list on the page where you’ll be able to:

• See the Model name and Provider of the models you have connected
• See the Health of the model
  • Healthy - The model is responding correctly and credentials are valid
  • Unhealthy - The model is not responding. Check credentials and provider status
• Select any model in the Models list to view detailed information:
  • Health status: Current health state (Healthy or Unhealthy)
  • Last checked: Timestamp of the most recent health check
  • Model details: Provider, model name, and model ID
• See who added the model to AI Studio and when

Select the three dot icon to the right of a model to Delete the model which will automatically remove it from your agents. For full details about configuring external models please visit our dev docs.

Best practices

Organizing models

Use clear naming conventions

• Include purpose in display name: "Palmyra X5 - Customer Support"
• Add environment indicators: "Claude - Production" vs "Claude - Testing"
• Indicate department if team-specific: "GPT-4 - Finance Team"

Group by use case

• Assign cost-effective models for high-volume, routine tasks
• Reserve advanced models for complex reasoning or specialized needs
• Keep experimental/testing models restricted to specific teams

Managing team access

Start restrictive, expand as needed

• Initially grant access to specific teams
• Monitor usage and costs
• Expand to "All Teams" once confident in the use case

Align access with capability

• Give specialized models (e.g., financial, medical) only to relevant teams
• Ensure teams have training on model capabilities before granting access

Regular audits

• Review team access quarterly
• Remove access for models no longer needed
• Update access when team members change

Credential management

Security best practices

• Use dedicated service accounts, not personal credentials
• Rotate credentials on a regular schedule
• Document credential ownership and renewal dates
• Store backup credentials securely outside AI Studio

Credential reuse

• Use the same credentials across multiple models from the same provider
• Reduces management overhead
• Easier to rotate when needed

Troubleshooting

Model shows as unhealthy

If your model shows an unhealthy status:

Need info here

Potential causes to check:

• Credentials may have expired
• Provider service may be experiencing downtime
• Network connectivity issues
• Model may have been deprecated by provider
• Regional restrictions or availability issues

Model doesn't appear when building agents

If you can't see a model when building an agent, check:

1. Team access restriction

• Check if the model is assigned to "Specific teams"
• Verify you're a member of an assigned team
• Contact your admin to request access if needed

2. Model health issue

• Check if the model shows as "Unhealthy" in the models list
• Contact your admin to resolve connection issues

3. Permissions

• Verify you have builder access in AI Studio
• Confirm your role permissions with your admin if needed

Permission errors when adding models

Error: "You don't have permission to add models"

Solution: Contact your org admin, IT admin, or user with AI Studio full access role to:

• Grant you appropriate permissions, or
• Add the model on your behalf

Credential errors

Error: "Invalid credentials" or "Authentication failed"

Solutions:

1. Verify credentials are copied correctly with no extra spaces
2. Check that credentials are still valid in your provider account
3. Confirm the AWS region matches your model deployment region
4. For Role ARN access, verify the role has necessary permissions
5. Try creating new credentials from your provider

Still having issues?

• Check your provider's service status page
• Review provider documentation for credential requirements
• Contact WRITER support with error messages and model details

FAQs

What providers are supported?

Currently supported providers:

• AWS Bedrock: Fully supported
• Hugging Face: Coming soon
• Nvidia: Coming soon

Can I use the same credentials for multiple models from the same provider?

Yes! When configuring a model, choose credentials you've already configured for that provider from the Credentials name list in the setup modal.

Benefits:

• Simpler credential management
• Easier credential rotation
• Consistent access patterns

How do I know which model to choose for my use case?

Consider these factors:

• For general tasks: Use balanced models like Palmyra X5
• For complex reasoning: Use advanced models like Palmyra X5 Thinking 
• For cost efficiency: Start with smaller models and upgrade if needed
• For specialized domains: Use industry-specific models (Palmyra Med, Palmyra Fin)
• For vision tasks: Use multimodal models like Palmyra Vision

💡 Tip: Consult our dev docs for detailed model comparisons and recommendations.

Do I need to configure credentials for WRITER models?

No. WRITER models (Palmyra family) are pre-configured and immediately available. You only need to configure credentials for external provider models.

Do I need separate credentials for each team?

No. Credentials are configured at the model level, not the team level. One set of credentials can be:

• Used by multiple models from the same provider
• Accessed by all teams that have access to those models
• Managed centrally by admins

Team access controls WHO can use the model, not the credentials themselves.

What happens if I'm not on a team with access to a specific model?

If a model is restricted to specific teams and you're not a member:

• You won't see the model when building agents
• You can request access from your admin
• Your admin can either:
  • Add you to an allowed team, or
  • Add your team to the model's access list

Can team members see the credentials?

No. Credentials are encrypted and only visible to users who configure them (admins). Team members with access can use the models but never see the underlying credentials.