davebcn87

davebcn87 / pi-autoresearch

Public

Autonomous experiment loop extension for pi

548
28
100% credibility
Found Mar 13, 2026 at 548 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

This repository provides an extension for an AI coding agent that automates iterative experiments to optimize project metrics like test speed or bundle size by trying changes, measuring outcomes, and retaining improvements.

How It Works

1
🔍 Discover the tool

You hear about a handy helper that lets your AI coding buddy automatically test and improve your project's speed or efficiency.

2
📥 Add it to your workspace

You simply add this optimization tool to your AI assistant's toolkit so it's ready to use.

3
🎯 Pick what to improve

You tell your AI the goal, like making tests run faster or builds smaller, and it sets everything up.

4
🚀 See experiments in action

Your AI starts trying changes, running tests, measuring results, and keeping only the winners—all on its own.

5
📊 Track the progress

Glance at the live status display or open the full results view anytime to see how much better things are getting.

✅ Celebrate faster results

Your project now performs better than before, with a complete record of successful tweaks you can build on.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 548 to 548 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is pi-autoresearch?

pi-autoresearch is a TypeScript extension for the pi coding agent that powers autonomous experimentation loops to optimize code performance metrics like test speed, bundle size, build times, or LLM training loss. You define a benchmark command and target metric once, then the agent autonomously edits code, commits changes, runs experiments, logs results, keeps improvements, and discards failures—repeating indefinitely. A persistent dashboard and status widget track progress, baselines, and deltas, surviving restarts via simple session files.

Why is it gaining traction?

It stands out by decoupling generic infrastructure (tools for running/timing any shell command and logging metrics) from domain-specific skills, letting you optimize anything from Lighthouse scores to materials discovery workflows without custom hacks. The always-visible widget and toggleable dashboard give instant feedback, while auto-resume across agent sessions keeps momentum without babysitting. Developers hook on the hands-free loop that turns vague perf ideas into tracked, branch-isolated experiments.

Who should use this?

Backend engineers tuning slow test suites or build pipelines, frontend devs shrinking bundles for production, and ML researchers accelerating training loops with custom val_bpb metrics. Ideal for anyone with repetitive optimization drudgery in pi, like autonomous agents iterating on configs for faster execution or better scores.

Verdict

With 548 stars and solid docs, it's mature enough for pi users chasing autonomous experimentation gains—install via `pi install` and kick off with `/skill:autoresearch-create`. The 1.0% credibility score flags early-stage risks like edge-case stability, so test on non-critical projects first; still, a smart add for perf obsessives.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.