The company says it uses GPT-6 Astra computer use to test application behavior in addition to reading pull-request code.
The described workflow can start a development server for a pull request, resize a browser to an iPhone-sized viewport, and simulate a user moving through the application. OpenAI gives an example of finding a button that disappears on a mobile view. Greptile also says its Knowledge Base indexes customer codebases and documentation, and exposes that context to Codex through a Model Context Protocol server.
Greptile CEO Daksh Gupta says thousands of Codex customers have connected to that knowledge base. The company tracks how often AI-generated code is reverted and says added context leads to fewer reverts, while also using Daybreak for customer defenses. The source does not provide revert rates, test coverage, detection accuracy, comparative benchmarks, pricing, or technical access details for the described workflows.