Chat Interface

The chat interface in Orchestra provides a single place to message back and forward between SLMs and kick off automated workflows which you've built.

Behind the Orchestra chat interface is our workflow model router. This router is a lightweight model which understands what workflows you have enabled and will route requests based on your prompt.

How it works

When workflows are saved, a description of what the workflow automates is generated. Based on what workflows are enabled, the router learns the capabilities, goals, and parameters of the selected workflows.

When a request is sent in the chat interface, the model determines whether this is a general question that the model you've selected can answer, or whether a workflow should be invoked. If it's a general question, the prompt is routed to the selected model and a response is provided. If the router determines a workflow should be called, it extracts the required parameters from the prompt for the specific workflow and then triggers the workflow to execute.

This means you can invoke all of your automations from the same place.

For example, from a single chat interface, a sales representative could take a company they've never heard of and automatically research the specific company and their industry, generate a sales talk track and key points to discuss tailored to this customer, add the customer to their CRM, draft an introductory email, and send it to the customer.

Models and Model Settings

Out of the box, you can select Virtuoso-Large, Virtuoso-Medium, Virtuoso-Small, and coder. You also have the ability to bring in other models.

You can set a system prompt to be passed to the model with each invokation, and can set hyperparameters such as max tokens, temperature, top p, frequency penalty, and presence penalty.

Workflow Selection

On the tools tab, you can select the workflows you want to be considered when invoking the model.

Files and images can also be uploaded through the chat.