SUPERWISE® automatically generates a dataset that records chat conversations upon agent creation. Leveraging the Elemeta meta features extractor, the platform extracts valuable features from the conversation text, providing insightful data analytics.

📘

Production interactions only

Please note that only interactions made against a deployed agent (whether via API or using our embedded iframe) are recorded. This means any interaction in our playground is not recorded by definition.

Once your agent interactions are recorded in a dataset, you can leverage the platform's capabilities to track, monitor, and visualize what’s going on:

• Gain Deep Insights: Explore advanced data exploration techniques - link to explore data guide.
• Monitor Effectively: Discover how to create data-driven monitoring policies - link to monitoring policy guide.
• Visualize for Clarity: Craft interactive dashboards to present your findings - link to dashboards creation guide.

Change your agent observability setting via UI

While observability is a key element in governing your agents, SUPERWISE® provides users with the option to opt in or out of the platform's observability service. This means that users can turn it off, instructing the platform not to collect any data about the production interactions of their agent. To do this, simply go to the "Settings" tab and turn off the observability section.

Change your agent observability setting via SDK

Observability is on by default, and can be set using the observability_enabled flag.

Either at creation time:

agent = sw.agent.create(f"my agent name", observability_enabled=True)

Or by updating an existing agent:

agent = sw.agent.update(agent_id=agent.id, observability_enabled=False)

Analyze user feedback

SUPERWISE® empowers you to track user feedback on chat responses. Each answer in our iframe comes with the option to provide a thumbs-up (positive) or thumbs-down (negative) rating, along with the ability to explain negative feedback.

🚧

Pay attention

SUPERWISE® only track the end user feedback received from the embedded app, and not from the playground

Analyzing Feedback is Simple:

1. Navigate to the Explore Screen.
2. Select the "Events" data source and filter by the event type "Agent feedback." If you have multiple agents, you can further refine the results by filtering the specific agent you're interested in.
3. Start analyzing!

For a deeper dive into using the query builder, refer to the documentation here.