Skip to content
Neural Daily

ToolsNews

Greptile uses GPT-6 Astra to click through pull requests

🇭🇺 Olvasd el magyarul →

Greptile uses GPT-6 Astra's computer use to run and test pull requests, not just read the code. It also hands its codebase knowledge base to Codex through an MCP server.

By: Mikola Máté Patrik · · 2 min read

OpenAI cover image with the OpenAI and Greptile logos beside a green square with a copper circuit pattern
Image: OpenAI · Press kit · Official OpenAI customer story image · Source

What was announced?

OpenAI published another customer story on Monday (OpenAI), this time about Greptile, which builds code review agents. The key point: with GPT-6 Astra's computer use, Greptile does not just read a pull request. It starts the application and walks through it like a user.

How does it work?

CEO Daksh Gupta's example is a button that disappears on an iPhone. Greptile starts a dev server for the PR, resizes the browser to an iPhone-sized viewport and clicks through the flow. The model inspects the screen and notices the missing button. Gupta says a QA firm testing every change is far too slow for how fast teams now ship code with these models.

The other piece is context. Greptile's Knowledge Base indexes a customer's codebase and documents how it works for both people and agents. Codex gets it through an MCP server, so it writes code that follows the customer's existing patterns. Greptile says thousands of customers have connected the two. It measures the impact by how often AI-written code gets reverted.

What to keep in mind

This is a customer story without hard numbers: it gives no bug detection rates and no revert reduction figures. The pattern is still useful. If PR validation includes a computer use agent actually clicking through the app, it catches bugs that reading the code misses. Sharing internal documentation with coding agents over MCP is also one of the cheapest ways to improve their output.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles