Skip to main content
Python package

goalseek

Local-first, git-backed research loops for coding agents.

goalseek orchestrates disciplined research loops for coding agents with manifests, mechanical verification, and a durable local artifact trail.

goalseek experiment lifecycle overview

The loop is simple on purpose

Baseline

Measure the current project without changing code and capture the retained metric.

Iterate

Plan, apply, verify, and commit one focused change per loop iteration.

Keep or Revert

Promote improvements and automatically roll back regressions with git-backed auditability.

Built for inspection, not magic

Git-backed

Every candidate change becomes a real commit, and rejected changes are reverted with history intact.

Local-first

Projects, manifests, logs, and experiment artifacts stay on disk in a layout you can inspect and version.

Provider-agnostic

Swap between Codex, Claude Code, Gemini, OpenCode, or a fake provider through a consistent loop engine.