Anthropic LLM analytics installation
- 1
Install the PostHog SDK
RequiredSetting up analytics starts with installing the PostHog SDK for your language. LLM analytics works best with our Python and Node SDKs.
- 2
Install the Anthropic SDK
RequiredInstall the Anthropic SDK:
Proxy noteThese SDKs do not proxy your calls. They only fire off an async call to PostHog in the background to send the data.
You can also use LLM analytics with other SDKs or our API, but you will need to capture the data in the right format. See the schema in the manual capture section for more details.
- 3
Initialize PostHog and the Anthropic wrapper
RequiredInitialize PostHog with your project API key and host from your project settings, then pass it to our Anthropic wrapper.
Note: This also works with the
AsyncAnthropicclient as well asAnthropicBedrock,AnthropicVertex, and the async versions of those. - 4
Call Anthropic LLMs
RequiredNow, when you use the Anthropic SDK to call LLMs, PostHog automatically captures an
$ai_generationevent.You can enrich the event with additional data such as the trace ID, distinct ID, custom properties, groups, and privacy mode options.
Notes:
- This also works when message streams are used (e.g.
stream=Trueorclient.messages.stream(...)). - If you want to capture LLM events anonymously, don't pass a distinct ID to the request. See our docs on anonymous vs identified events to learn more.
You can expect captured
$ai_generationevents to have the following properties:Property Description $ai_modelThe specific model, like gpt-5-miniorclaude-4-sonnet$ai_latencyThe latency of the LLM call in seconds $ai_toolsTools and functions available to the LLM $ai_inputList of messages sent to the LLM $ai_input_tokensThe number of tokens in the input (often found in response.usage) $ai_output_choicesList of response choices from the LLM $ai_output_tokensThe number of tokens in the output (often found in response.usage)$ai_total_cost_usdThe total cost in USD (input + output) ... See full list of properties - This also works when message streams are used (e.g.

