Agents in the Wild

👋 Overview

This repository contains the source code for insights.logicstar.ai.
We track all GitHub pull requests created since 2025-05-15 and analyze them to identify autonomous code agents.

Pull requests are classified using the following rules:

Human: All PRs that do not match any known agent pattern.
OpenAI Codex: PRs where the head branch starts with codex/.
Google Jules: PRs where the first commit is authored by google-labs-jules[bot].
GitHub Copilot: PRs where the head branch starts with copilot/.
Devin: PRs authored by devin-ai-integration[bot].
Cursor Agent: PRs where the head branch starts with cursor/.
Claude Code: PRs where the first commit is authored by claude or the head branch starts with claude/.
OpenHands: PRs where the first commit is authored by openhands.
Codegen: PRs authored by codegen-sh.
Tembo: PRs authored by tembo-io.
Cosine: PRs where the head branch starts with cosine/

⚙️ System Architecture

The system consists of:

A Next.js frontend for querying the database and visualizing insights.
A Python backend for scraping GitHub pull requests and generating insights.

🚀 Installation

While we host a live version at insights.logicstar.ai, you can also deploy the system locally.

💿 Database

Ensure you have access to a PostgreSQL database.

💻 Frontend

Make sure you have npm and node installed. Then run:

npm install
npm run build

Configure the environment via frontend/.env.local:

DATABASE_URL=postgres://<user>:<password>@localhost:5433/<db>

To start the frontend:

npm run start

⚙️ Backend

Install the backend package in development mode:

pip3 install -e backend

You should now have access to the aitw CLI tool.

Set the following environment variables:

POSTGRES_CONNECT="dbname=<dbname> user=<user> password=<passwd> host=<host> port=<port>"
GOOGLE_APPLICATION_CREDENTIALS="<path_to_gcp_app_creds>"

We use GCP for logging scraping worker activity. You can omit the credentials if you're not using GCP logging.

To start a scraping worker, ensure the environment is configured and run:

aitw scrape worker --token <GH_TOKEN>

To backfill all PRs since 2025-05-15:

aitw scrape manager backfill

To derive insights from the current database:

aitw insights hourly
aitw insights daily

To update with new PRs created or closed since the last update:

aitw scrape manager update

We recommend running both the update and insights commands regularly (e.g., hourly).

📝 Changelog

2025-07-10: Added “Cite Us” section, released dataset, and added OpenHands, Cosine, Claude Code, and Codegen agents. 2025-07-08: Initial release of the dashboard.

logic-star-ai/insights