The AI Agent Connection enables you to define intelligent agents that run directly in your BlueConic environment. These agents deliver insights about your customer data and tenant configuration through scheduled or ad hoc runs. AI agents can automatically send results such as alerts, audits, or analyses—directly to your inbox, eliminating the need for external systems.
Use the AI Agent Connection to:
Automate tenant configuration audits.
Detect anomalies in profile data, connection runs, and notebook runs.
Send daily insight summaries or alerts.
Note: The AI Agent Connection is in public preview. Some functionality is still in development and may change. You can try the feature at no additional cost and share feedback with BlueConic.
Before you begin
Before creating an AI Agent Connection, ensure the following requirements are met:
Generative AI features are enabled for your tenant.
You have the Edit AI Agent permission in your role.
Users without this permission have read-only access to existing AI Agent instances and run logs. They cannot create or edit connections.
Create an AI Agent Connection
Via the BlueConic navigation, select Connections.
Click Add Connections.
Search or select AI Agent Connection via the pop-up window.
Enter the name and metadata details for the connection.
Click Save.
Configure the AI Agent Connection
In the Prompt field, enter instructions for the AI agent.
Request insights based on tenant configuration or profile data.
Include conditional logic to email results (to, cc, bcc).
Select Start from template to load a sample prompt or write your own.
(Optional) Enable Send email notifications to receive updates about agent run status (success or failure).
This setting only controls run status alerts. It does not send the agent’s insights by email.
To email insights, include instructions in the agent prompt, for example:
Send results as an HTML email to [address], with subject [subject line].
Turn the Connection ON.
Click Save.
Choose how to run the agent:
Now: Run the agent immediately.
On a schedule: Schedule the agent to run automatically (maximum once per day).
Click Save.
Profile Property Insights
Tip: Use the built-in example prompts in the configuration window to quickly create a working setup.
Email example results
Inactive Listener Alert
Email example results
Prompting tips
Well-structured prompts lead to more reliable, explainable, and actionable AI Agent results. Use the following best practices when writing prompts.
Explicitly define the objects to analyze
Where possible, list the exact objects you want analyze. Reference objects by their UUIDs rather than by name to avoid guesswork and ambiguity.
Constrain the number of objects retrieved
For example, to analyze a segment, instruct the AI to retrieve a bounded number of profiles (e.g., the first 25-50 profiles per segment). This helps keep the analysis grounded in the actual data the model is working with.
Require traceability
Ask the AI to include a short “inputs used” section in its response, such as:
Segment UUIDs used
Number of profiles retrieved per segment
Profile properties used
This makes validation easier and helps identify any mismatches.
Add guardrails to avoid assumptions
For example, instruct the AI to explicitly state when a segment or property cannot be found, rather than attempting to infer or fill in gaps.
Clarify aggregation vs sample-based results
When prompts ask for totals or aggregates, it’s important to distinguish between full aggregations and calculations based on a limited set of retrieved profiles. Unless explicitly stated otherwise, treat such results as sample-based.
Privacy and security
AI agents operate only within your tenant and can access configuration and customer data through the public REST API read endpoints.
All agent activity is logged under [email protected].
Agents can send emails only to users with a BlueConic user account.
The run history log captures the full prompt, result message, and any tool calls used during execution.
Thresholds and limitations
API access | Not all configuration object types and settings are accessible to agents. To see what data is available, refer to the BlueConic REST API documentation. |
Run frequency | Agents can run at most once per day. |
LLM calls | Each run supports up to 50 LLM calls. Each LLM call supports up to approximately 800,000 characters, including the request and response message. |
Key-value storage | Agents can use key-value storage, limited to 50 keys per agent, with each key supporting up to 5,000 characters. This storage is scoped to the specific connection instance and not shared. |
Next steps
Review your AI Agent run logs to verify results and email output.