Single-binary Go tools for running LLMs in production

Four command-line tools, one job each: a coding harness, a research agent, an invocation-log analyser and a telemetry router. Pure-function cores, pluggable backends, secrets never in config.

Read the docs Source on GitHub 
# a bare invocation lands in read-only planning mode$ ./stirrup harness --prompt "Outline a fix for the failing test"

Tools

stirrupthe tack that steadies the rider

A production-grade coding harness for safe autonomous operation — five deterministic safety rings, a pure-function core, a declarative RunConfig.

Documentation →
chironthe wise centaur, the teacher

Drives a Deep Research agent end-to-end from the CLI and emits a cited, structured Markdown report. Research-only: no shell, no edits.

Documentation →
dressagethe discipline of refined training

Fetches provider invocation logs, normalises them, groups them into conversations, and generates a self-contained HTML report.

Documentation →
steeplechasethe long course, the fences

A lightweight OTLP telemetry router — metrics, logs, and traces received once and fanned out to stdout, other OTLP backends, or MQTT.

Documentation →

Design principles

Pure-function cores
Every tool is a pure core behind pluggable backends. Components swap without modifying the loop.
Secrets never in config
Credentials come from the environment. Secret leakage through logs is structurally impossible.
One auditable binary
A single static Go binary per tool. Minimal dependencies, OTel-native, air-gapped-friendly.