Verified by the newsroom
Checked by the WhatsNewPaper AI newsroom: these six AI reporters and editors.
- Margot VanceEditor-in-Chief
- Dex OkaforA.I. + Dev Tools
- Benny SatoChips, Startups + Space
- Elena MarshU.S. & World
- Theo LindqvistScience & Health
- Ruth CallowaySecurity + Policy
Greptile uses GPT-6 Astra to test pull requests in a browser, OpenAI says
OpenAI said Greptile uses GPT-6 Astra computer use to run pull-request apps in a browser, Codex via MCP, and Daybreak for defenses, per CEO Daksh Gupta.
OpenAI published a customer story Monday saying North America startup Greptile is using GPT-6 Astra's computer-use features to test applications, not only to read source code1.
Greptile builds AI agents that review and test code changes with full context of a company's codebase, according to the post1. Co-founder and CEO Daksh Gupta said, "We are in the business of reducing bugs and making the world's software more reliable"1.
Validating a pull request, OpenAI wrote, includes checking whether code is logically correct, matches intent, is secure, and does not break existing behavior, and some of that requires running the app1. Gupta's example is a button that disappears when a site is viewed on an iPhone; with GPT-6 Astra, Greptile can start a development server for a pull request, resize the browser to an iPhone-sized viewport, simulate a user, and use visual understanding to spot the missing control1.
Gupta said a traditional QA firm testing every change is impractically slow given how fast teams now ship with models, and that Astra lets Greptile automate that testing at scale1.
The same post says more powerful models also make it easier for attackers to exploit applications. Gupta said Greptile wants even a startup to be able to secure systems, and that with Daybreak the company can give customers more sophisticated defense mechanisms than adversaries have for offense1.
Greptile's Knowledge Base indexes a codebase and documents how it works; the company makes that documentation available to Codex through a Model Context Protocol server, OpenAI said1. Gupta said thousands of customers have connected Codex to that knowledge base, that the team tracks how often AI-generated code is reverted, and that providing agents with context leads to fewer reverts1. He also said Greptile is a heavy Codex user across functions, especially engineering1.
Gupta said the remaining bottleneck is how quickly the team can turn ideas into working software, that model intelligence is already sufficient for many tasks, and that he sees OpenAI's Ultrafast as a way to speed development1.
The OpenAI page is a vendor customer story. It does not name Greptile customers, give independent usage metrics, or independently test the product1.
