GitHub Repository

Lightweight local OpenTelemetry backend — OTLP receiver, Parquet storage, DuckDB queries in a single binary

2 starsGo

Ducktel – observability when the consumer is an AI agent, not a human

Name: Ducktel – observability when the consumer is an AI agent, not a human
Availability: InStock
Author: djhope99

by djhope99·Mar 9, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●●BangerZero to OneBig Brain

Agent-first observability with SQL output instead of human dashboards.

Strengths

•OTLP ingestion with Parquet storage means no database to manage.
•DuckDB embedded SQL is a language LLMs already understand well.
•Saved queries let agents run diagnostic checklists autonomously.

Weaknesses

•No metrics support mentioned, only traces and logs.
•Zero stars and forks suggests very early, untested in production.

Post Description

Receives OTLP, writes Parquet, exposes DuckDB SQL via CLI. No dashboards, no query builders, no alerting engine. An LLM agent shells out to ducktel query, gets JSON back, and does the reasoning itself. The thesis: everything observability platforms built above ingest+store was scaffolding for human cognition. When the consumer is an LLM, the scaffolding is just overhead.