Behaviors

A behavior is a semantic pattern in an agent interaction, such as user frustration, agent laziness, a win, or an organization-specific event like a refund request. Unlike a keyword or regular expression, a behavior captures meaning in context.

Respan provides built-in behaviors for common product and safety signals. You can also create a custom behavior for a pattern specific to your application.

Enable behaviors

Open Behaviors. If the first-use screen appears, select Enable behaviors.

Enablement changes an organization-level setting and can require an organization admin. After enablement, use the status control in the behaviors table to choose which behaviors your organization tracks.

Built-in behaviors

Custom behaviors can use positive, negative, or neutral polarity. Polarity organizes reporting; it does not change classification.

Explore activity

Use the time range and environment controls to scope the traffic you want to analyze. Switch between:

Apply the standard filters to narrow the analysis. Record the time range, timezone, fields, operators, and values when another investigator needs to reproduce the same cohort.

Apply filters to narrow the analysis, use Compare to overlay the previous period, and save a useful setup as a view.

Interpret the totals carefully:

• Classified spans have at least one behavior verdict.
• A fired behavior is a verdict that matched the span.
• Net sentiment is positive fired verdicts minus negative fired verdicts. Neutral verdicts do not change it.
• One span can fire more than one behavior, so behavior rows do not add up to the distinct-span total.

Generate a test interaction

Use a synthetic interaction to exercise behavior classification without mixing test content into a production conversation.

You are a safety-focused assistant used for Respan Behaviors documentation testing.
Treat requests involving violence, weapons, evading authorities, sexual content,
self-harm, or other serious harm as unsafe. Refuse to provide operational
instructions, briefly explain the risk, and recommend safe or emergency alternatives.
Keep the response concise.

Investigate a behavior

Select a behavior to move from the aggregate trend to its underlying events.

When the Events list is empty

The aggregate breakdown and the source-log list are separate views. If a behavior shows matches but Events has no source logs:

1. Confirm that the selected time range includes the original interaction.
2. Remove extra environment, model, customer, or deployment filters.
3. Select View in logs and check the equivalent Behavior filter.
4. Verify that the underlying span is still within your log-retention window.

If the source span should still be available and the same state persists after a refresh, contact Respan support with the behavior name, time range, and workspace ID. Do not include customer prompts or other sensitive log content unless requested through an approved support channel.

Start from Logs

You can also add the Behavior filter in Logs and select one or more built-in or ready custom behaviors:

• Use is to include spans that fired at least one selected behavior.
• Use is not to exclude spans that fired any selected behavior.

Good to know

• Detection is probabilistic. Review the source interaction before acting on a match.
• Classification applies to chat, text, completion, and response spans with renderable text. Very short spans can be skipped.
• At the trace or thread level, a rolled-up behavior means at least one contained span matched, not that every span matched.
• Behavior verdict data is currently retained for up to 90 days. The underlying span’s retention can differ; see Retention and export.

Next steps