Agent Detection

TL;DR: Colter classifies AI traffic by provider, agent name, and interaction type. Lens uses that data for traffic analytics, and Check uses it to score ecosystem coverage.

Overview

Colter maintains a registry of known AI agents and crawler variants across major providers. Incoming traffic is matched by User-Agent, request patterns, and provider-specific heuristics.

Core Classifications

Class	Meaning	Examples
user_action	Shopping agents acting for a user	ChatGPT-User, Operator, Claude-User, Gemini
search	Crawlers used for AI search or retrieval	GPTBot, ClaudeBot, PerplexityBot, Googlebot
training	Crawlers used for model training or corpus expansion	Google-Extended, Bytespider, Meta-ExternalAgent

Major Providers

Provider	Example agents	Primary use
OpenAI	ChatGPT-User, GPTBot, OAI-SearchBot, Operator	Shopping, search, training
Google	Google-Extended, Googlebot, Gemini	Search, training, shopping
Anthropic	ClaudeBot, Claude-User, Claude-SearchBot	Search, shopping
Microsoft	Copilot, BingBot	Search, shopping
Meta	Meta-ExternalAgent, FacebookBot	Search, training
Other providers	Applebot, Amazonbot, PerplexityBot, Bytespider, more	Mixed

How Lens Uses Detection

Every Lens event carries:

• agent_name
• agent_class
• agent_provider

That makes it easy to filter traffic, build funnels, and alert on new or suspicious agents. When Lens sees an unknown pattern worth review, it can raise a new_agent alert.

Signature Verification

Some agents publish RFC 9421 HTTP Message Signatures. Lens verifies those signatures when available and reports verified versus unverified traffic. Unsigned traffic can still be classified; signatures are an extra trust signal, not a requirement.

How Check Uses Detection

Check maps agent families to ecosystem coverage:

• Google ecosystem coverage depends on UCP.
• OpenAI ecosystem coverage depends on ACP.

That mapping feeds the ecosystem score and the final readiness verdict.

Next Steps

• Lens
• Check
• Scoring Model
• MCP Tools