Monitor

With Inconvo you can easily monitor your users’ conversations with your agents.

Conversations

Every conversation created through the API will be visible in the conversations table.

Conversation Feedback

Conversations containing answers with feedback are given a feedback rating of Positive, Mixed or Negative.

Feedback Rating Ranges:

• Positive: averageFeedback >= 0.6 (60% or higher)
• Negative: averageFeedback <= 0.3 (30% or lower)
• Mixed: 0.3 < averageFeedback < 0.6 (between 30% and 60%)

Examples:

• Conversation with all positive feedback: avg = 1.0 → Positive
• Conversation with all negative feedback: avg = 0.0 → Negative
• Conversation with 2 positive, 1 negative: avg = (1+1+0)/3 = 0.67 → Positive
• Conversation with 1 positive, 1 negative: avg = (1+0)/2 = 0.5 → Mixed

Filtering Conversations

Conversations can be filtered by feedback type:

• Negative Feedback: Shows conversations containing at least one answer with negative feedback.
• Positive Feedback: Shows conversations containing at least one answer with positive feedback.

Conversation Details

Clicking on row within the conversations table allows you to view the details of a particular conversation.

Click the trace button on any response to view the steps your agent took to generate it.

Debugging with Traces

Traces show every step your agent took to generate a response — the tools it called, the queries it ran, and the data it returned. When a response is incorrect, traces help you identify the root cause.

Common patterns

Pattern	What you see in the trace	Fix
Wrong table	Agent queries users instead of customers	Add a table prompt clarifying terminology
Missing filter	Query returns data across all tenants	Add a context filter on the table
Ambiguous column	Agent picks created_at instead of last_order_at	Add a table prompt explaining which column to use for the concept
Wrong join	Agent joins through an unrelated table	Toggle the incorrect relation to Off

Walkthrough

1. Open a conversation from the monitoring table
2. Click the Trace button on the response you want to debug
3. Review the tool calls — check which tables and columns were selected
4. If the query is wrong, go to your semantic model and add context (table prompts, computed columns, or relation changes)
5. Re-run the query to confirm the fix

Feedback Loop

Combine monitoring with the feedback API to continuously improve agent quality:

1. Collect feedback — Use the feedback API to log positive or negative ratings on responses from your users
2. Filter by feedback — Use the monitoring table to filter conversations with negative feedback
3. Debug with traces — Open traces on negative responses to identify what went wrong
4. Fix the semantic model — Update table prompts, context filters, or relations as needed
5. Validate — Re-run failing queries to confirm they now return the correct results