pi-mono/packages/ai/docs/telemetry-schema.md

Pi AI Telemetry Schema

Schema version: 1

pi.ai.request

One logical request to an AI provider

• Parents: root or any caller span
• Default status: ok
• Error when: The operation throws or returns an error result

Start attributes

Name	Type	Required	Values	Notes	Description
pi.ai.operation	string	yes	stream, fetch_deferred, cancel_deferred, generate_images		Logical provider operation
pi.ai.provider	string	yes			Selected provider id
pi.ai.model	string	yes			Requested model id
pi.ai.api	string	yes			Provider API id
pi.ai.streaming	boolean	yes			Whether this operation returns a stream
pi.ai.deferred	boolean	no			Whether the operation requests or participates in deferred execution

End attributes

All end attributes are optional completion enrichment.

Name	Type	Values	Notes	Description
pi.ai.response.model	string			Concrete response model
pi.ai.response.id	string		high cardinality	Provider response id
pi.ai.response.stop_reason	string	stop, length, tool_use, error, aborted, deferred		Normalized terminal response reason
pi.ai.http.status_code	number			Final HTTP status
pi.ai.usage.input_tokens	number			Reported input tokens
pi.ai.usage.output_tokens	number			Reported output tokens
pi.ai.usage.cache_read_tokens	number			Reported cache-read tokens
pi.ai.usage.cache_write_tokens	number			Reported cache-write tokens
pi.ai.usage.reasoning_tokens	number			Reported reasoning tokens
pi.ai.usage.total_tokens	number			Reported total tokens
pi.ai.usage.cost	number			Reported total cost
pi.ai.stream.chunk_count	number			Streamed update chunk count
pi.ai.stream.time_to_first_chunk_ms	number			Elapsed milliseconds to first update chunk
pi.ai.error.type	string		low cardinality	Provider or transport error class

Events

No declared span events.