FAQs and Troubleshooting

See Pricing for the most up-to-date information.

We are taking an AI Constellation model approach, which defines which AI models and prompt strategies your AI Assistant will use. They allow you to select from different model types to choose the appropriate LLM provider and logic that best fits your application's needs. Constellations allow your Assistant to leverage a number of available agents to best meet your application's specific needs. This approach will also improve latency.

While AI Assistants are in developer preview, there are feature limitations and system limits in place. Learn more about current limitations in Limitations.

If you want to exceed these limitations, please reach out to our team at twilioalpha@twilio.com.

AI Assistants can speak in multiple languages, however we've not tested or optimized for multilingual support yet. It is on our roadmap to test and make improvements to multilingual support. In the meantime English is the recommended language for the configuration of your Assistants.

If you want your test your Assistant in other languages you can define this in the Personality Prompt:

Ensure that your responses are always in the same language the customer uses.

All API requests will still end up in the Conversation History in near real-time, so you can see what knowledge and tools were called there. Conversation history is available via API and in the Console.

Every Assistant comes with a REST API that you can use to trigger a message. Aside from a body you can pass in a session_id of your choice (used to maintain message history) and an identity. Learn more in the Channels documentation.

You can also use Twilio Quick Deploy to quickly connect your AI Assistant to various channels.

Tools differ from Knowledge in that Tools accomplish specific tasks (for example, looking up the weather or finding directions to a location), whereas Knowledge entries are unstructured data sources that the AI Assistant can reference to ask additional questions (such as your company's policies or documentation).

Use Tools when you're looking to take action or fetch specific data. Use Knowledge when you want to interpret an unstructured data set (that is, answering questions or referencing information).

If your AI Assistant is providing unexpected responses or isn't using Tools or Knowledge as intended, this troubleshooting guide can help you identify and resolve common issues. Follow the steps below to check and address potential problems with your Assistant's performance.

Issue: You need to understand how the Assistant processes requests and where it might be deviating from expected behavior.

Solution: Use the Simulator to test various queries and review the Assistant's decision-making process. This lets you see which configuration, Tools, or Knowledge entries the Assistant uses and provides insight into potential misconfigurations.

Tip: Take note of specific responses that seem off and check if the Assistant is defaulting to an unintended Tool or Knowledge source. Adjust descriptions or configurations as needed.

Issue: The Assistant isn't using specific Tools or Knowledge sources as expected.

Solution: Check the configurations for each Tool and Knowledge source:

• For Tools: Ensure each Tool is correctly defined with the proper attributes (name, description, input schema, HTTP method, etc.). Verify that the Assistant has access to all necessary endpoints.
• For Knowledge sources: Confirm that the Knowledge entries are indexed and that the Assistant has completed processing them. If a Knowledge entry is outdated, refresh it manually or enable auto-update.

Tip: Check that the descriptions for Tools and Knowledge sources are clear and help the Assistant determine when to use them. Ambiguous descriptions may lead to unexpected usage. For example, here's a draft Knowledge source:

Name: Employee Handbook

Issue: The Assistant's tone or style doesn't match your configuration, or it isn't providing responses aligned with its personality prompt.

Solution: Open the Assistant's personality prompt configuration and ensure it reflects the desired tone, voice, and purpose. If you recently made updates to the prompt, test to confirm they were saved properly.

Tip: Use concise and clear language in your prompt, and avoid overly complex instructions, as these can lead to ambiguous responses. For example:

Issue: The Assistant isn't remembering customer interactions or using previous data effectively.

Solution: Confirm that Customer Memory is enabled and configured correctly. Check that the integration with Twilio Segment is active and that data is being stored and retrieved properly.

Tip: If customer data is not updating as expected, check the AI Perception Engine settings to ensure conversation data is being captured in Segment.

Issue: API calls aren't producing expected results, or responses contain errors.

Solution: Inspect recent API calls to and from your Assistant for errors or unexpected data. Review logs for any failed API calls or unusual response codes.

Tip: Verify that authentication and access tokens are valid and correctly configured. If you're using custom endpoints, confirm they're accessible from Twilio and returning expected data.

Issue: After following the troubleshooting steps, your Assistant still isn't functioning as expected.

Solution: If issues persist, reach out to twilioalpha@twilio.com for assistance. Provide detailed examples of unexpected behavior, relevant logs, and any troubleshooting steps you've taken. Please provide your Twilio Account SID, and the specific AI Assistant ID (found under "My Assistants") and session ID (found in the Conversation History) if the issue pertains to a specific conversation.

Tip: Gather as much information as possible before contacting support, as this will help expedite the resolution process.