Req plugin that automatically captures $ai_generation events for LLMs.

It tries to extract as much information as possible from both requests and responses. Currently, it works best with the following APIs:

• OpenAI (Responses)
• OpenAI (Chat Completions)
• Anthropic (Create Message)
• Gemini (generateContent)

Usage

Just add it to your Req client before making a call:

Req.new()
|> PostHog.Integrations.LLMAnalytics.Req.attach()
|> Req.post!(url: "https://api.openai.com/v1/responses", json: %{model: "gpt-5-mini", input: "Who are you?"})

Optionally, start a new span beforehand to add additional properties to the event:

PostHog.LLMAnalytics.start_span(%{"$ai_span_name": "OpenAI Request"})
Req.post!(client, url: "https://api.openai.com/v1/responses", json: ...)

Integrating with InstructorLite

InstructorLite built-in adapters allow customizing the HTTP client using the http_client option. Define a wrapper module like this:

defmodule ReqWithLLMAnalytics do
  def post(url, opts) do
    Req.new(url: url)
    |> PostHog.Integrations.LLMAnalytics.Req.attach()
    |> Req.post(opts)
  end
end

Then pass this module as the http_client option in adapter_context. Optionally, start a span beforehand!

PostHog.LLMAnalytics.start_span(%{"$ai_span_name": "LLM Call"})

InstructorLite.instruct(%{
    input: [%{role: "user", content: "John is 25yo"}],
    model: "gpt-4o-mini"
  },
  response_model: %{name: :string, age: :integer},
  adapter: InstructorLite.Adapters.OpenAI,
  adapter_context: [
    api_key: "my-secret-key",
    http_client: ReqWithLLMAnalytics
  ]
)
{:ok, %{name: "John", age: 25}}