Observability built for agents
Tael is an AI-agent-native observability platform. A single tael binary ingests OpenTelemetry traces, logs, and metrics over standard OTLP and exposes a CLI-first, JSON-native query surface — designed so that agents like Claude Code can inspect their own telemetry, not just humans staring at dashboards.
There are no dashboards to click through and no browser to open. Every command returns structured JSON by default (--format table when a human is looking), so an agent can run tael query traces --status error --format json, parse the result, and decide what to investigate next.
See it in action
Server up, an agent reporting, and the whole investigation — query, correlate, annotate — from the terminal:
Why Tael
- CLI-first, JSON-native. Every command prints structured JSON by default. An agent parses the output directly — no HTML scraping, no screenshots.
- Single binary.
cargo install tael-cligives youtael.tael serveruns OTLP ingest, storage, and the REST API in one process — no Docker, no cluster. - OTLP-native. Ingests standard OpenTelemetry over gRPC on
:4317. No proprietary SDK; point any OTel exporter at Tael and you're done. - Agent-native query surface.
tael summarize,tael anomalies, andtael correlateanswer the questions an agent actually asks — "what's broken, how bad, and what's the whole story for this trace?" — in one shot. - LLM-trace aware.
gen_ai.*semantic conventions are flattened into typed columns: model, token counts, cost, time-to-first-token. Prompts and completions are stored as content-addressed blobs and full-text searchable. - Trace comments. Agents annotate traces with their findings, so investigations persist across sessions and build an audit trail.
A 60-second tour
Start the server, point your app at it, and ask what happened:
# 1. Run the server (OTLP on :4317, REST API on :7701)
tael serve
# 2. Point any OpenTelemetry app at it
export OTEL_EXPORTER_OTLP_ENDPOINT=http://127.0.0.1:4317
export OTEL_SERVICE_NAME=my-agent
# 3. Ask what happened
tael services
tael query traces --status error --last 1h
tael get trace <trace-id>
tael summarize --last 15mWhere to go next
- Installation — install the binary and start the server.
- Quickstart — instrument an app and run your first queries end to end.
- Wide Events — the one practice that makes everything else work.
- Debugging with an Agent — the order of operations for diagnosing a live issue.
- CLI Reference — every command, flag, and example.
