Switchfrog
Switchfrog is a session-level detection tool that identifies when AI browser agents like Claude, Codex, or Cursor take over a user's authenticated session. It reads the session itself rather than just traffic patterns, distinguishing human from agent activity in milliseconds and enabling real-time responses. The platform provides a dashboard showing which accounts agents use and what they do, with configurable actions to allow, block, or bill for agent usage.
- Cybersecurity
- Artificial Intelligence
- Software Only
Funding
Funding not disclosed
Founders
Founder details are not available yet.
Product
Problem
Millions of people are using AI browser agents like Claude Cowork, Codex, and OpenClaw that operate through users' real browsers and authenticated accounts. Because these agents use legitimate logged-in sessions, they bypass traditional bot detection and look like ordinary product usage, leaving companies blind to agent-driven activity on their platforms.
Solution
Switchfrog detects when an AI agent takes over a user's session, showing which accounts agents use and what they do in real time. The tool reads the session itself rather than just traffic patterns, allowing it to distinguish a person from an agent operating on the same account. It identifies sessions as human, agent, or unknown, typically within milliseconds, via a one-line code installation. Companies can then choose how to respond—allowing the agent, blocking it, or billing the customer for an agent seat. The dashboard displays session events such as when a specific agent like "Claude in Chrome" takes over, giving teams immediate visibility into agent behavior.
Target Audience
Product teams at software companies whose customers use AI browser agents to interact with their services, including SaaS platforms, developer tools, and consumer applications that need visibility into automated usage of authenticated accounts.
Features
- Session-level AI agent detection that identifies specific agents including Claude, Codex, Cursor, Grok, Gemini, Perplexity, Copilot, and OpenClaw in Chrome and other browsers
- Real-time classification of sessions as human, agent, or unknown with result aggregation (e.g., 76% human, 19% agent, 5% unknown)
- One-line code installation for rapid deployment into existing applications
- Millisecond-level detection latency for immediate response capability
- Configurable response actions: allow, block, or bill for agent usage on a per-session basis
- Session event timeline showing human-to-agent takeover transitions on the same account