Greptile builds AI agents that review and test code changes with full knowledge of the codebase. “We are in the business of reducing bugs and making the world’s software more reliable,” says Daksh Gupta, Greptile co-founder and CEO.
Why they run applications, not just read code
Gupta argues that validating a change often requires running the application, because reading source code alone can miss behavior-driven problems. To scale that testing, Greptile leverages the computer-usage capabilities of GPT‑6 Astra.
He gives the example of a button that disappears when a website is viewed on an iPhone. Using GPT‑6 Astra, Greptile can start a development server for a pull request, resize the browser to an iPhone viewport, and simulate a user navigating the app. The model’s visual understanding lets it inspect the screen and detect the hidden button.
Daksh notes that having an external QA firm test every change is impractically slow given modern release cadences, whereas GPT‑6 Astra allows this class of application testing to be automated at scale.
Security: Daybreak and protecting smaller teams
While more capable models help automations, Daksh warns they also make it easier for attackers to exploit applications. To address that risk, Greptile uses Daybreak to strengthen defenses in a way they describe as safe for customers.
“Our goal is to make it very easy for even a startup to secure their systems,” Daksh says, adding that Daybreak helps customers deploy defense mechanisms that are, in his view, more sophisticated than the offensive tools adversaries might use.
Knowledge Base, Codex and fewer reverts
Greptile’s Knowledge Base indexes a company’s codebase and documents how it works in a format readable by people and agents. That documentation is made available to Codex via a Model Context Protocol (MCP) server.
This setup lets Codex produce code that follows a customer’s existing patterns. According to Daksh, thousands of customers have connected Codex to Greptile’s knowledge base. To measure impact, the team tracks how often AI-generated code is reverted; providing agents with context leads to fewer reverts, saving customers time and inference costs.
Those savings can be redirected to new work rather than reworking the same changes.
Greptile uses Codex internally too
Greptile is also a heavy internal user of Codex, applying it to build their own software and agents. Daksh says the company has enormous usage across functions, especially engineering.
Bottlenecks and long-term goals
For Greptile, the main bottleneck is how quickly the team can turn ideas into working software, not model intelligence. Daksh believes model capabilities are already sufficient for many tasks and sees Ultrafast as a way to accelerate development.
In the longer term, Greptile aims to make changing and shipping critical enterprise software as easy as shipping a side project, with OpenAI helping build the required validation.
Conclusion
By combining GPT‑6 Astra’s runtime and visual capabilities, Daybreak’s security checks, and Codex’s contextual code generation (via MCP), Greptile aims to automate code and application testing to reduce bugs, lower reverts, and speed up software delivery while maintaining stronger defenses.
