Generations

Generations are events that capture LLM calls and their responses. They represent interactions and conversations with an AI model. Generations are tracked as $ai_generation events and can be used to create and visualize insights just like other PostHog events.

View recent AI generation events in the Activity tab

The LLM analytics > Generations tab displays a list of generations, along with a preview of key autocaptured properties. You can filter and search for generations by various properties.

Preview and filter AI generations in the Generations tab

What does each generation capture?

A generation event records the AI model’s inputs, generated output, and additional metadata – like token usage, latency, and cost – for each LLM call.

PostHog automatically logs and displays the generation and its data within a conversation view for contextual debugging and analysis. You can also view the raw JSON payload.

You can expect each generation to have the following properties (in addition to the default event properties):

Property	Description
`$ai_model`	The specific model, like `gpt-5-mini` or `claude-4-sonnet`
`$ai_latency`	The latency of the LLM call in seconds
`$ai_tools`	Tools and functions available to the LLM
`$ai_input`	List of messages sent to the LLM
`$ai_input_tokens`	The number of tokens in the input (often found in response.usage)
`$ai_output_choices`	List of response choices from the LLM
`$ai_output_tokens`	The number of tokens in the output (often found in `response.usage`)
`$ai_total_cost_usd`	The total cost in USD (input + output)
...	See full list of properties

When calling LLMs with our SDK wrappers, you can also enrich the $ai_generation event with your own custom properties and PostHog attributes like groups and distinct IDs for identified users.

response = client.responses.create(
    model="gpt-5-mini",
    input=[
        {"role": "user", "content": "Tell me a fun fact about hedgehogs"}
    ],
    posthog_distinct_id="user_123", # optional
    posthog_trace_id="trace_123", # optional
    posthog_properties={"custom_property": "value"}, # optional
    posthog_groups={"company": "company_id_in_your_db"},  # optional 
    posthog_privacy_mode=False # optional
)

Generations are nested under spans and traces.

A trace is the top-level entity that groups all related LLM operations, including spans and generations, together.

Spans are individual operations within a trace. Some spans represent generations, which are also uniquely identified using the $ai_span_id property. However, most spans track other types of LLM operations such as tool calls, RAG retrieval, data processing, and more.

Generations and spans are nested within a trace

Event properties

A generation is a single call to an LLM.

Event name: $ai_generation

Core properties

Property	Description
`$ai_trace_id`	The trace ID (a UUID to group AI events) like `conversation_id` Must contain only letters, numbers, and special characters: `-`, `_`, `~`, `.`, `@`, `(`, `)`, `!`, `'`, `:`, `\|` Example: `d9222e05-8708-41b8-98ea-d4a21849e761`
`$ai_span_id`	(Optional) Unique identifier for this generation
`$ai_span_name`	(Optional) Name given to this generation Example: `summarize_text`
`$ai_parent_id`	(Optional) Parent span ID for tree view grouping
`$ai_model`	The model used Example: `gpt-5-mini`
`$ai_provider`	The LLM provider Example: `openai`, `anthropic`, `gemini`
`$ai_input`	List of messages sent to the LLM Example: `[{"role": "user", "content": [{"type": "text", "text": "What's in this image?"}, {"type": "image", "image": "https://example.com/image.jpg"}, {"type": "function", "function": {"name": "get_weather", "arguments": {"location": "San Francisco"}}}]}]`
`$ai_input_tokens`	The number of tokens in the input (often found in response.usage)
`$ai_output_choices`	List of response choices from the LLM Example: `[{"role": "assistant", "content": [{"type": "text", "text": "I can see a hedgehog in the image."}, {"type": "function", "function": {"name": "get_weather", "arguments": {"location": "San Francisco"}}}]}]`
`$ai_output_tokens`	The number of tokens in the output (often found in response.usage)
`$ai_latency`	(Optional) The latency of the LLM call in seconds
`$ai_http_status`	(Optional) The HTTP status code of the response
`$ai_base_url`	(Optional) The base URL of the LLM provider Example: `https://api.openai.com/v1`
`$ai_request_url`	(Optional) The full URL of the request made to the LLM API Example: `https://api.openai.com/v1/chat/completions`
`$ai_is_error`	(Optional) Boolean to indicate if the request was an error
`$ai_error`	(Optional) The error message or object

Cost properties

Cost properties are optional as we can automatically calculate them from model and token counts. If you want, you can provide your own cost property instead.

Property	Description
`$ai_input_cost_usd`	(Optional) The cost in USD of the input tokens
`$ai_output_cost_usd`	(Optional) The cost in USD of the output tokens
`$ai_total_cost_usd`	(Optional) The total cost in USD (input + output)

Cache properties

Property	Description
`$ai_cache_read_input_tokens`	(Optional) Number of tokens read from cache
`$ai_cache_creation_input_tokens`	(Optional) Number of tokens written to cache (Anthropic-specific)

Model parameters

Property	Description
`$ai_temperature`	(Optional) Temperature parameter used in the LLM request
`$ai_stream`	(Optional) Whether the response was streamed
`$ai_max_tokens`	(Optional) Maximum tokens setting for the LLM response
`$ai_tools`	(Optional) Tools/functions available to the LLM Example: `[{"type": "function", "function": {"name": "get_weather", "parameters": {...}}}]`

Example API call

Terminal

curl -X POST "https://us.i.posthog.com/i/v0/e/" \
     -H "Content-Type: application/json" \
     -d '{
         "api_key": "<ph_project_api_key>",
         "event": "$ai_generation",
         "properties": {
             "distinct_id": "user_123",
             "$ai_trace_id": "d9222e05-8708-41b8-98ea-d4a21849e761",
             "$ai_model": "gpt-4o",
             "$ai_provider": "openai",
             "$ai_input": [{"role": "user", "content": [{"type": "text", "text": "Analyze this data and suggest improvements"}]}],
             "$ai_input_tokens": 150,
             "$ai_output_choices": [{"role": "assistant", "content": [{"type": "text", "text": "Based on the data, here are my suggestions..."}]}],
             "$ai_output_tokens": 280,
             "$ai_latency": 2.45,
             "$ai_http_status": 200,
             "$ai_base_url": "https://api.openai.com/v1",
             "$ai_request_url": "https://api.openai.com/v1/chat/completions",
             "$ai_is_error": false,
             "$ai_temperature": 0.7,
             "$ai_stream": false,
             "$ai_max_tokens": 500,
             "$ai_tools": [{"type": "function", "function": {"name": "analyze_data", "description": "Analyzes data and provides insights", "parameters": {"type": "object", "properties": {"data_type": {"type": "string"}}}}}],
             "$ai_cache_read_input_tokens": 50,
             "$ai_span_name": "data_analysis_chat"
         },
         "timestamp": "2025-01-30T12:00:00Z"
     }'

What does each generation capture?

Event properties

Core properties

Cost properties

Cache properties

Model parameters

Example API call

Questions? Ask Max AI.

Community questions

Was this page useful?

Spans

Edwin Lim

Generations

What does each generation capture?

How are generations, traces, and spans related?

Event properties

Core properties

Cost properties

Cache properties

Model parameters

Example API call

Questions? Ask Max AI.

Community questions

Was this page useful?

Spans