This project provides a command-line interface (CLI) tool, Oracle, designed to enhance AI agent interactions by allowing users to bundle prompts with context from specified files. It targets developers and power users seeking to integrate AI models like GPT-5.1 Pro into automated workflows, offering a structured way to provide rich context for more accurate and grounded AI responses.

How It Works

Oracle operates using two distinct engines: an API engine that interfaces directly with OpenAI's API (requiring an OPENAI_API_KEY) and a browser engine that automates ChatGPT within Chrome. This dual-engine approach allows flexibility, with the API engine being stable and preferred, while the browser engine offers a keyless alternative for Pro accounts, albeit experimental. The tool bundles user-provided prompts with specified files, supporting glob patterns and exclusions, into a markdown package that AI models can process and cite.

Quick Start & Requirements

• Primary install/run command: npx -y @steipete/oracle
• Prerequisites: OPENAI_API_KEY environment variable for the API engine. For the browser engine: an OpenAI Pro account and macOS with Chrome installed.
• Links: docs/chromium-forks.md for details on using Chromium-based browsers.

Highlighted Details

• Bundles prompts and files into a single markdown package for AI consumption, enabling model citations.
• Supports flexible file selection via glob patterns and exclusion rules (!).
• Offers robust handling for GPT-5.1 Pro, including long-running sessions (up to an hour) with reconnection and token/cost tracking.
• Provides a unified user experience across both API and browser engines.
• Includes file safety features like per-file token accounting and size guards, with an option to report usage.

Maintenance & Community

No specific details on maintainers, community channels, or roadmap were found in the provided text.

Licensing & Compatibility

The specific open-source license was not mentioned in the provided text.

Limitations & Caveats

The browser engine is considered experimental and is currently limited to macOS with Chrome. Windows and Linux browser support is under development. The browser engine also requires an OpenAI Pro account. For stability and broader compatibility, the API engine is recommended when an API key is available.