See which AI agents
follow your links.
Claude Code, Cursor, LangChain agents, GPT researcher bots — they're already following links you share. You just can't see them.
Until now.
No spam. No pitch. We'll reach out when we're ready for you.
The problem
Every analytics tool is built for humans.
Click = person opened a browser = you track their country, device, referrer. That model made sense in 2020.
Today, Claude Code, Cursor, LangChain executors, and AutoGen agents are all browsing the web on behalf of their users. They follow links. They fetch resources. They navigate docs. All of it shows up as Unknown / desktop / direct.
← these are AI agents. you just can't tell.
The solution
AgentLink.
Every AWSYS.CO short link now detects when it's followed by an AI agent and captures which agent, which model, and which framework — automatically.
No instrumentation. No SDK changes. No changes to your agents. Your existing short links become agent-aware the moment you opt in.
Built different
Everything LangSmith tracks happens inside your agent. AgentLink tracks what your agent does out in the world.
Agent identity
Which agent? Which model? Which framework? Claude Code, LangChain, Cursor, Crew — named and attributed.
Zero instrumentation
Works at the link layer. Nothing to install, no SDK to wrap, no changes to your agents or pipelines.
Link-flow analytics
Agent vs human ratio. Session timelines. Which docs your agents actually read. A dataset that doesn't exist anywhere else.
First-class attribution
Agents that want precise attribution add one header. Everything else is automatic from UA patterns and MCP headers.
Complements your stack
LangSmith and Helicone instrument the agent process. AgentLink instruments the agent's environment. They answer different questions.
MCP-native
AWSYS.CO already has an MCP server. AgentLink closes the loop — agents can query their own link stats via MCP tools.
How detection works
Three layers, combined for confident classification.
User-Agent registry
A curated, continuously-updated registry of known agent UA patterns — Claude Code, Cursor, LangChain, Crew, AutoGen, LlamaIndex, Playwright, and 20+ more.
MCP header detection
Requests coming through MCP clients carry identifiable headers (x-mcp-session-id, x-mcp-tool-call-id) that unambiguously identify them as agent-originated.
Self-identification (optional)
Agents that want first-class attribution add a single header. No other changes required.
x-awsys-agent: {"agentId":"my-agent","modelName":"claude-sonnet-4-6","taskId":"task-123"}
FAQ
Is this available now?
Yes — live on Pro and Builder plans today. Every click on your short links is classified as agent or human automatically. New frameworks are added to the registry on a rolling basis; tell us if yours isn't recognized.
Do I need to change anything in my agents?
No. Detection is automatic. For precise attribution (agent ID, task ID, model name), your agent can add a single header — but that's optional, not required.
What if my agents don't identify themselves?
We classify traffic using User-Agent patterns, MCP headers, and request heuristics. Most popular frameworks are already covered. Framework maintainers can contact us to get added with first-class recognition.
Agents can spoof their User-Agent — doesn't that break detection?
We use multi-signal classification. UA alone isn't enough — we also look at MCP headers, Accept-Language presence, request timing, and other heuristics. Confidence scores are surfaced so you know when a detection is probabilistic vs certain.
How does this relate to LangSmith, Helicone, or Langfuse?
Those tools instrument the agent process — LLM calls, costs, latency, evals. AgentLink instruments the agent's environment — what links it follows, what resources it reads. They're complementary. Most teams will use both.
Does this work for existing AWSYS.CO links?
Yes. All existing short links automatically gain agent-aware tracking when you opt in. No new links required.
Stay informed
We add new agents and frameworks to the registry on a rolling basis. Drop your email and we'll let you know when yours is recognized — or when a new detection method goes live.
✓ You're on the list.
We'll reach out personally when we're ready for you.
No spam. No pitch. Unsubscribe any time.