Agent Intelligence helps you understand how AI agents operate inside your application by surfacing detected agentic sessions and the details behind them. The Sessions view focuses on individual sessions, helping you review which surfaces agents reached and inspect the signals Mosaic used to classify each session.
This page explains the available session data and how Agent Intelligence turns collected signals into session-level data. For integration guidance, start with Integration options or go directly to one of these guides:
- Integrate using Cloudflare Workers
- Integrate using Akamai EdgeWorkers
- Integrate using Google Tag Manager
- Integrate using the Platform SDK directly
The following terms describe how Agent Intelligence groups and classifies activity in the Sessions view.
| Term | Definition |
|---|---|
| Agent platform | A specific tool that executes actions on a user's behalf, such as ChatGPT Agent, Manus AI, or Operator. |
| Agent category | The class an agent platform belongs to, assigned automatically by Mosaic. Categories are not customer-configurable and are extended as new platform types emerge. Current categories include: • AI-native agentic browser (examples: Browserbase, Dia, Arc) • AI-augmented mainstream browser (examples: Chrome with Gemini, Edge with Copilot) • OS-level / Computer-use agent (examples: Claude Computer Use, Operator, Devin) • Headless / developer framework (examples: Playwright, Puppeteer, Selenium, LangChain) • Browser extension agent (examples: Riskified, HubSpot, Grammarly) |
| Browsing session | A continuous span of activity from a single agent or user, bounded by a period of inactivity. This is the unit Agent Intelligence uses to group actions and pages into one visit. |
| Unclassified | Sessions confidently detected as agentic but not yet attributed to a named agent platform. |
At a high level, Agent Intelligence follows this flow:
- The Platform SDK is loaded into the application, either directly in the app or through an edge/CDN integration.
- As the application runs in the browser, the SDK captures browser and action signals in the background.
- Those signals are sent to Mosaic for backend processing.
- Mosaic correlates the activity, analyzes agentic patterns, and turns the processed results into session-level data.
- Agent Intelligence then exposes that data as session volume, per-session summary fields, and deeper session and path details.
Agent Intelligence exposes session data at three levels: aggregate volume, per-session summary fields, and additional session and path details for each individual session.
The aggregated session data is available over the selected date range and is split into two series:
- Detected platforms — sessions attributed to a named agent platform.
- Unclassified — sessions confidently detected as agentic but not yet attributed to a named agent platform.
Comparing these two series shows both how much agentic traffic your application receives and how much of it Mosaic can already attribute to a known platform.
For each session, Agent Intelligence provides the following summary fields:
| Column | Description |
|---|---|
| Session ID | Unique identifier of the session. |
| Activity Surface | Application surface where the session was detected, such as login or /. |
| Start Time | When the session started. |
| Agent Platform | Agent platform detected for the session, such as ChatGPT Agent or Manus AI. Sessions that couldn't be attributed appear as Unclassified. |
| IP Address | IP address observed for the session (IPv4 or IPv6). |
| Country | Country resolved from the IP address. |
| OS | Operating system reported for the session. |
| Browser | Browser or runtime used by the agent, such as Chrome or Electron. |
| Origin Type | Where the session ran: Cloud/Hosted for agents running on hosted infrastructure, End-User Device for agents running on the user's own device. |
| Anonymization | Anonymization technique detected on the connection. Shows — when none is detected. |
| Self-Declared | Verified when the agent declared its identity and Mosaic confirmed the declaration, Observed when the session was classified from observed signals only. |
For any individual session, Agent Intelligence also makes additional data available beyond the summary fields above.
Session-level details include:
- The detected Agent category for the session's platform.
- Network and Location context for the session.
- Session duration, which shows the total recorded length of the session.
- Actor type, which indicates whether the activity was attributed to an Agent, a Human, or both in the session.
Path-level details summarize the activity recorded inside the session:
- Total duration: total duration of the observed path in the session.
- Pages: number of pages visited in the session.
- Interactions: number of recorded interactions in the session.
Path data lists the pages visited in order. Each step shows the page, its detection status, the time spent on it, and the actor type recorded for that step.
Together, these sections let you move from "an agent was here" to what it actually did: which pages it reached, how long it stayed, and how much it interacted.
Choose the integration that best fits your project: