Cluster

Detecting the AI Stack

The AI stack a company runs leaves fingerprints: SDK calls, client libraries, and named API hostnames. This cluster covers how to read each layer of it from a domain alone.

5 questions about detecting the ai stack

• Can you detect LLM observability tools like LangSmith or Helicone?

  You can sometimes detect LangSmith or Helicone, but it is the hardest layer to read from outside. Both usually run server-side, so they surface only when a client SDK, a proxy hostname, or a bundled reference leaks into the front end.

• Can you tell if a company uses Anthropic or OpenAI?

  You can sometimes tell whether a company uses Anthropic or OpenAI by fingerprinting client-side SDK code, model ID strings, and API calls. Most apps proxy the model server-side, which hides the provider, and Bassethound reports that gap honestly.

• How do you detect a company's vector store?

  You detect a vector store by fingerprinting the client libraries, package names, and API hostnames a site loads. Pinecone, Weaviate, Qdrant, Chroma, and Milvus each leave a distinct trail.

• How do you detect LangChain or LlamaIndex on a website?

  You detect LangChain or LlamaIndex by fingerprinting their package namespaces, client APIs, and telemetry endpoints in the code a site ships. Both are backend frameworks, so a server-only deployment can stay invisible.

• How do you detect which AI models a company uses?

  You detect which AI models a company uses by fingerprinting the provider SDKs, package names, and API hostnames its front end loads and calls. Anthropic, OpenAI, Gemini, Cohere, and Mistral each leave a recognizable trail.