Agent Toolkit lets you give an AI agent Pendo capabilities, so it can guide, support, and act on behalf of your users. To do this, it draws on the product usage data Pendo already collects and, if you turn it on, on browser errors captured from the visitor's session.
This article describes what data Agent Toolkit uses and collects, how it's stored, and the privacy and security measures in place.
Note: Agent Toolkit is available in open beta. It's part of Pendo for Agents, which brings Agent Toolkit and Agent Analytics together. For more information, see Overview of Pendo for Agents.
Prerequisites
To capture console and network errors, you must be on Pendo Web SDK version 2.338.0 or later. The rest of Agent Toolkit works with earlier SDK versions.
What data is used
Agent Toolkit doesn't introduce a new stream of behavioral data. It uses the in-app analytics Pendo already collects for your application:
- Pages and Features
- Guides
- Frustration signals
- Network activity
Which of these an agent can use depends on the capabilities you turn on for it. For example, session context draws on recent Pages, Features, and frustration signals; workflow nudges and proactive support draw on network activity and frustration signals. For more information, see Overview of Agent Toolkit.
What new data is collected
Agent Toolkit can capture console errors and network errors from the visitor's browser, so your agent can respond to problems a visitor is running into. Previously, capturing browser errors required Session Replay with an active recording. With Agent Toolkit, Pendo can capture these errors without Session Replay.
Error capture is off by default. To turn it on for every web app an agent can access, use Collect all console and network errors on the agent's Session context tab. To turn it on for a single web app, go to Settings > Subscription settings > Applications, select the app, and use the same setting in Basic settings. The app-level setting appears only for web apps that don't use Session Replay. Until you turn it on, Agent Toolkit collects no console or network error data.
Agent Toolkit respects your privacy and exclusion settings. Pendo drops all data from visitors and accounts marked as Do Not Process before it's stored, including console and network errors and session context.
Where data is stored
Visitor activity and error data collected through Agent Toolkit are stored using the same infrastructure and security practices as other Pendo product data.
Pendo uses Google Cloud Platform (GCP) to host customer data in a secure, multi-tenant environment. GCP services meet industry standards for security and compliance, including SOC 2 Type II and ISO 27001. Customer data collected through Agent Toolkit is never commingled with data from other subscriptions.
Console and network error data is stored with the rest of your application's event data and deleted on your subscription's data retention schedule. For session context, Pendo also keeps a visitor's most recent errors, truncated, for 15 minutes.
For details about data storage, encryption, and compliance certifications, see Data collection and compliance and Security and privacy at Pendo.
How Pendo uses AI in Agent Toolkit
Two Agent Toolkit capabilities use large language models (LLMs):
- Session context. Summarizes a visitor's recent activity before it's shared with your agent.
- Frustration signals. Uses a filter to decide which signals are worth acting on.
Pendo stores the inputs and outputs of these models and logs them in Agent Analytics to improve the quality of the features. The inputs are the product usage data Pendo already collects for your application, described in the What data is used section. Neither Pendo nor its AI providers use your data to train the large language models behind these features.
For the list of AI providers Pendo uses, see Artificial intelligence (AI) at Pendo.
Preventing sensitive data capture
Agent Analytics redaction doesn't apply to console and network error data. Instead, the Pendo Web SDK masks this data before sending it, the same way it masks Session Replay console and network logs: it masks email addresses, IP addresses, Social Security numbers, credit card numbers, and URLs; masks values under JSON keys such as password, token, auth, key, phone, address, and ssn; and truncates captured bodies to 1,000 characters.
To redact more, add your own logic with the networkLogs.captureRequestBody and networkLogs.captureResponseBody callbacks in pendo.initialize. Pendo doesn't apply further redaction after the data arrives, so error data is stored as the SDK sent it.
For general information about Pendo's security and data privacy practices, see our Trust Center.