Interacting with Your Models

Once you've built and deployed AI models, Navigator and Companion provide multiple ways to interact with them. Whether you're testing a flow locally, chatting with a deployed model, or leveraging advanced reasoning capabilities, this guide covers everything you need to know about engaging with your AI creations.

Overview

Your AI models can be accessed through two main applications:

• Navigator: The development environment where you build, test, and iterate on your AI workflows using the Preview Tab and other features.
• Companion: The private chat interface where you interact with deployed models in a conversational format.

Both applications support advanced features like reasoning mode for models that have this capability, giving you transparency into how your AI arrives at its conclusions.

Testing Models in Navigator

Navigator provides several ways to test and interact with your models during development, allowing you to iterate quickly without deploying to production.

Running Flows Locally

The fastest way to test your model is by running it locally in Navigator:

1. Open your project in the Canvas view
2. Configure your elements and connections
3. Click the Run button to start your flow
4. Navigator will automatically provision your local device for testing

When running locally, Navigator uses your computer's resources to power the AI model. Ensure your device meets the system requirements for the model you're testing.

Using the Preview Tab

For LLM-based flows, the Preview Tab offers an interactive chat interface directly within Navigator:

• After running your flow, Navigator automatically switches to the Preview Tab
• Chat with your model in real-time to test responses
• Adjust element settings without returning to the Canvas
• Iterate quickly on prompts and configurations

Iterative Testing Workflow

1. Run your flow and switch to Preview
2. Test your chatbot with various prompts
3. Click the dropdown next to Run and select Stop to pause
4. Modify settings in the configuration panel
5. Click Run again to test your changes

Interacting with Models in Companion

Companion is your dedicated chat interface for interacting with deployed AI models. It provides a clean, conversational experience that makes using your custom AI as simple as chatting with a colleague.

Selecting a Model

When you open Companion, you'll see all available models in the left panel:

• Default Local Model: Companion automatically selects a model optimized for your device's hardware.
• Deployed Models: Any models you've deployed from Navigator appear as separate options.
• Organization Models: Models deployed by your organization or clusters you belong to are also accessible.

Starting a Conversation

Interacting with your models in Companion is straightforward:

1. Select a model from the left panel (look for the green status indicator)
2. Type your question or prompt in the chat input
3. Click the blue arrow to send your message
4. Watch as your model generates a response in real-time

Managing Conversations

Companion helps you stay organized with robust conversation management:

• Create New Threads: Click the Chat icon to start a fresh conversation while preserving previous ones.
• Search History: Use the Search bar to find previous conversations by content or topic.
• Switch Between Models: Each model maintains separate conversation histories for easy context switching.

Enhanced Chat Features

Companion offers several features to enhance your AI interactions:

Web Search

Enable web search to let your model access current information from the internet:

• Toggle the web search icon in the chat input area
• Useful for questions about recent events or real-time data
• Requires an active internet connection

Speech-to-Text

Use voice input for hands-free interaction:

• Click the microphone icon to activate voice input
• Speak naturally and watch your words appear as text
• Ideal for longer prompts or accessibility needs

File Generation

Companion can generate code and files directly in the chat:

• Request HTML, Python, JavaScript, or other file types
• Generated files appear inline in the conversation
• Preview, edit, and save files from the Files panel

Reasoning Mode

Reasoning Mode is a powerful feature available in both Navigator and Companion that provides transparency into how your AI model thinks through problems. When enabled, you can see the step-by-step thought process your model uses to arrive at its conclusions.

Benefits of Reasoning Mode

Enabling Reasoning Mode offers several advantages:

• Transparency: Understand exactly how your AI reached its conclusions, building trust in the outputs.
• Debugging: Identify where the model's logic may have gone astray for complex queries.
• Learning: See the model's problem-solving approach to understand its capabilities and limitations.
• Verification: Confirm that the model is using appropriate logic for your specific use case.

Using Reasoning in Navigator

When testing models with reasoning capabilities in Navigator's Preview Tab:

1. Run your flow and navigate to the Preview tab
2. Look for the Reasoning toggle in the chat interface
3. Enable Reasoning Mode before sending your prompt
4. The model's response will include its thought process alongside the final answer

Using Reasoning in Companion

To enable Reasoning Mode in Companion:

1. Open Companion and select a reasoning-capable model
2. Look for the Reasoning Mode icon at the bottom of the chat input (next to Web Search and Speech-to-Text)
3. Click to toggle Reasoning Mode on
4. Send your message and observe the model's thinking process

Understanding Reasoning Output

When Reasoning Mode is active, responses include two components:

• Thinking Process: A detailed breakdown of how the model approached the problem, including intermediate steps, considerations, and logic chains.
• Final Response: The model's conclusion or answer based on its reasoning process.

Reasoning Mode may increase response time as the model performs more thorough analysis. For simple queries, you may prefer to keep Reasoning Mode disabled for faster responses.

Multi-Modal Interactions

Some models support multi-modal inputs, allowing you to include images alongside text in your conversations. This enables powerful use cases like image analysis, visual question answering, and document interpretation.

Multi-Modal capabilities are only available for vision-enabled models. Use the filters within the LLM element settings to see what models support Multi-Modal interaction. 

 

Using Images in Companion

When using a multi-modal model in Companion, you can attach images to your messages:

1. Select a multi-modal model from your deployed models
2. Click the attachment icon in the chat input area
3. Select an image file from your Photos app or take a new photo with your devices camera
4. Add your text prompt describing what you'd like the model to do with the image
5. Send your message and receive a response based on both the image and text

Using Images in Navigator

When testing multi-modal flows in Navigator's Preview Tab:

• Ensure your flow uses a multi-modal LLM 
• Run your flow and navigate to the Preview Tab
• Use the attachment option to include images with your test prompts
• Iterate on your prompts to optimize image understanding for your use case

Tips for Better Image Interactions

• Image Quality: Use clear, well-lit images for best results. Blurry or low-resolution images may produce less accurate responses.
• Be Specific: Provide clear instructions about what aspect of the image you want analyzed.
• Single Focus: For complex images, ask about one element at a time rather than requesting comprehensive analysis in a single prompt.
• File Size: Large image files may take longer to process. Consider resizing very large images before uploading.

Best Practices for Model Interaction

Effective Prompting

• Be Specific: Clear, detailed prompts yield better responses than vague questions.
• Provide Context: Give your model relevant background information for complex queries.
• Iterate: Use follow-up questions to refine and improve responses.
• Use Examples: When asking for specific formats, provide examples of what you're looking for.

Staying Organized

• Create separate conversation threads for different topics or projects
• Use descriptive first messages that will serve as conversation titles
• Regularly review and clean up old conversations to keep Companion organized

Choosing the Right Model

• Select models sized appropriately for your hardware
• Use reasoning models for complex analytical tasks
• Consider specialized models (code, vision) for domain-specific tasks
• Test multiple models to find the best fit for your use case

Troubleshooting

Slow Responses

• Close resource-intensive applications to free up system memory
• Consider using a smaller or more quantized model version
• Disable Reasoning Mode for faster responses on simple queries
• Ensure your device meets minimum hardware requirements

Reasoning Not Available

• Verify that you're using a model with reasoning support
• Use the filters on the LLM element settings in Navigator to find available models
• Ensure the Reasoning Mode toggle is properly enabled

Model Connection Issues

• Verify that your deployment is active in Navigator
• Ensure both Navigator and Companion use the same account credentials
• Ensure the email you are logged in under is apart of the organization running the deployed models
• Restart Companion to refresh the model list
• Check that your cluster is online and properly configured