Telemetry

Harnest uses the same logging and tracing API with ADK and LangGraph.

Configure one sink

Send logs and traces to one OTLP collector using environment variables.

Configure multiple sinks

Fan out signals with lifecycle-owned exporters.

Threadify

Link sessions to business threads and filter framework noise.

Add logs and spans

tools/search_catalog.py

from harnest.logging import get_logger
from harnest.tracing import span, traced

logger = get_logger("catalog").bind(component="search")


@traced("catalog.lookup", attributes={"catalog.kind": "products"})
def lookup(query: str):
    with span("catalog.rank", candidate_count=12):
        logger.info("catalog.lookup.completed", result_count=3)
        return ["one", "two", "three"]

Logs emitted inside a span include its trace and span IDs. Structured logs are JSON on stderr by default; trace export starts when you configure a sink.

Data boundary

Recorded by default	Not recorded by default
Operation names, outcomes, and timing	Prompts and responses
Correlated failures	Credentials and headers
Bounded playground traces in process memory	Raw session IDs and payloads

Runtime mutations emit audit events after commit as structured logs. They reach sinks configured for logs. CLI and compiler file changes are outside this boundary.

Overview

Get started

Guides

Configure one sink

Configure multiple sinks

Threadify

Add logs and spans

Data boundary

Configure one sink

Configure multiple sinks

Threadify

​Add logs and spans

​Data boundary

Add logs and spans

Data boundary