Greptile began at a hackathon in SoMa with this simple idea - could we teach LLMs to answer hard questions in plain English about large, real-world codebases, the way a senior engineer would? After less than a year, and with nearly 400 software teams using our API, chatbot, and integrations, we are thrilled to announce our $4.1M seed round led by Initialized Capital, with participation from angels including Rich Aberman (WePay) and JJ Fliegelman (WayUp), both former YC visiting partners.
Codebase Context is All You Need
We believe that software engineering is difficult not because of the complexity of the incremental code being written, but rather the difficulty of understanding the wider context of the codebase. The same is true for AI developer tools, from simple Q&A bots to …
Greptile began at a hackathon in SoMa with this simple idea - could we teach LLMs to answer hard questions in plain English about large, real-world codebases, the way a senior engineer would? After less than a year, and with nearly 400 software teams using our API, chatbot, and integrations, we are thrilled to announce our $4.1M seed round led by Initialized Capital, with participation from angels including Rich Aberman (WePay) and JJ Fliegelman (WayUp), both former YC visiting partners.
Codebase Context is All You Need
We believe that software engineering is difficult not because of the complexity of the incremental code being written, but rather the difficulty of understanding the wider context of the codebase. The same is true for AI developer tools, from simple Q&A bots to AI software engineering agents. Given the right context, sufficiently sophisticated LLMs can execute virtually any codegen and comprehension task.
Existing tools struggle to answer the types of nuanced, non-obvious questions a professional developer would have about their company codebase. We developed novel ways to index and search large codebases that mirror how experienced engineers do it: first by developing a general map/graph of how the codebase works, then following references – “command+clicking” – to find the code that answers the question. We offer this “codebase-optimized RAG” (Retrieval-Augmented Generation) as a chatbot that developers can use to chat with any repo, and also as an API so they can integrate this AI expert into their dev tooling. This API can support everything from reviewing merge requests, to diagnosing bugs from Sentry alerts, to updating internal documentation.
Our RAG is hyper-optimized for codebases. While most tools of this kind view codebases as lists of files, we view them as graphs, and index them in a way that maintains an understanding of how the different parts of the codebase work together. This allows Greptile to easily answer questions in plain English other tools struggle with since they require a nuanced understanding of how the codebase works.
Unlike most other functions, software teams are all unique in how they operate. Anywhere from 0-100% of their tooling might be built in-house, and they have huge variation in the processes that make up their software development lifecycle.
By providing our AI codebase expert as an API, software teams can integrate it exactly how they want to. Some examples of internal dev tools teams are building on Greptile:
- Enriching Sentry alerts with codebase context.
- Reviewing PRs, pointing out things like relevant abstraction and potential security concerns.
- Auto-updating documentation when new code is committed.
- Adding instructions and considerations to every JIRA ticket/Linear Issue.
Quote from Daksh about why this is the best solution: “No two software teams run the same way. So instead of forcing them to buy a suite of opinionated SaaS apps from different vendors to add AI to their dev lifecycle, we provide a single API platform that can be easily configured into any type of AI dev tool.”
Having a single AI expert is also advantageous because it creates cohesion: the bot that is answering codebase questions in Slack is the very same one that is reviewing PRs, commenting on design docs, and enriching JIRA tickets.
Co-founders Soohoon, Vaishant, and Daksh began building Greptile during a hackathon in San Francisco. The team got their first paying customers within a week of the event and were already up to 100 users before joining Y Combinator.
“Code intelligence is most valuable when it’s directly in the context where you need it – not stuck in a chat interface,” says Brett Gibson, managing partner at Initialized. “Greptile’s is not only the best intelligence/search we’ve seen, but it’s also packaged for developers to build in the contexts that make the most sense. We’ll likely see entirely new applications arise from this.”
The team is small, highly technical, always shipping, and actively recruiting. If you’re a programmer interested in building tools to help developers get more out of their SDLC, consider joining us! We’re in-person in San Francisco, serving over 400 software teams, and growing very fast. Email me to learn more -> daksh@greptile.com
Demo the product directly on our website: www.greptile.com