Observability

These settings control the logs and traces of tiny-harness. A log is one line for one event. A trace shows each step of a task and how long it took: the model calls, the tool calls and the retries.

tiny-harness removes every secret from logs and traces.

Example

[o11y]
log_level = "INFO"
otlp_endpoint = "https://otel.example.com"
service_name = "acme-support-agent"

[o11y] settings

Setting	Default
log_level	"INFO"
otlp_endpoint	none
service_name	"tiny-harness"
trace_file	none
langfuse_host	"https://cloud.langfuse.com"

log_level

The least important log level that tiny-harness writes.

• Default: "INFO"
• Values: "DEBUG", "INFO", "WARNING", "ERROR"
• Logs are JSON lines on standard output. When the TUI runs the agent in its own process, the logs go to tiny-harness.log beside the store.

otlp_endpoint

The address of an OpenTelemetry collector. tiny-harness sends traces to it over OTLP/HTTP. Most monitoring systems accept OTLP.

• Default: none. tiny-harness sends no traces.

service_name

The name of this program in your traces.

• Default: "tiny-harness"
• Set it when: you run more than one agent and want to tell them apart.

trace_file

A file where tiny-harness also writes each trace step as one JSON line.

• Default: none.
• Set it when: you want traces without a collector, for example during development. The file grows until you rotate it.

langfuse_host

The address of your Langfuse server.

• Default: "https://cloud.langfuse.com"
• tiny-harness sends traces to Langfuse only when you set both LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY.

Health

GET /_monitor on the server returns the tasks that run now and their states. It updates every heartbeat.interval.