Creating Assistants
Configuring the Assistant
The configuration panel, located on the left side of the interface, allows users to define the core attributes and behavior of the assistant. The following settings are available:
Name
The display name of the assistant. Choose a clear and descriptive title so that shared users can easily understand its purpose.
Description
A short summary of what the assistant does. A well-written description helps colleagues quickly identify when and how to use the assistant.
Instructions
This field defines the assistant’s personality, goals, and operational behavior. You may include:
Purpose and intended use
Tone, voice, and communication style
Formatting rules for responses
Step-by-step reasoning instructions
Any functional requirements for outputs
These instructions shape how the AI will behave in all interactions.
Conversation Starters
Users can add sample prompts that appear when the assistant is opened. These help new users understand the assistant’s capabilities and provide a quick way to begin interacting with it.
AI Models
Users may select the AI model used to generate responses. Response quality, reasoning ability, and cost may vary depending on the model chosen.
Capabilities
Users can enable the assistant’s data and reasoning options by toggling capabilities:
World Knowledge
Uses the model’s built-in, internet-scale parametric knowledge.
Ideal for general information and broad tasks.
Enterprise Knowledge (Add Files)
Assistant responds strictly based on the content of attached files.
Users can upload and associate up to 50 files with the assistant.
Ensures grounded, traceable, and organization-specific outputs.
World Knowledge with Web Search
Allows the assistant to retrieve relevant, real-time information from the web.
Useful for tasks requiring up-to-date facts or external references.
Integrated Actions (Apps & Services)
Assistants can be extended with powerful operational abilities by enabling integrations. When activated, the assistant can read, write, and take actions within connected applications—turning it into an execution tool, not just a conversational one.
Available integrations include:
Outlook Email – Read inbox, draft emails, send emails, organize messages.
Outlook Calendar – Check availability, schedule meetings, modify events.
Gmail – Read and send emails, manage threads, handle labels.
Google Calendar – Create events, update schedules, check availability.
Slack – Send messages, fetch conversations, post updates to channels.
HubSpot – Access CRM records, update contacts or deals, log activities.
Google Sheets – Read and update spreadsheets, append rows, modify data.
Microsoft Dynamics 365 CRM – Retrieve and update CRM entries, manage leads, access account data.
These integrations allow your assistant to perform real-world workflows—such as booking meetings, sending emails, updating CRMs, or managing project trackers—directly through the chat interface.
Users may select the capabilities that best match the assistant’s intended use case.
Testing the Created Assistant
The preview window on the right allows users to interact directly with the assistant they are building. Through live testing, users can:
Verify behavior
Check response quality
Confirm adherence to instructions
Evaluate grounding and accuracy (where applicable)
Testing ensures that the assistant performs as expected before it is used or shared.
Refining the Assistant Configuration
Based on the test interactions, users may return to the configuration panel to adjust:
Instructions
Capabilities
Attached files
Model selection
Conversation starters
Description or naming
Users can repeat this cycle—adjust, test, refine—until the assistant fully meets the target requirements.
Last updated