We led Greptile's seed round in June 2023. Greptile is building an API for LLM-powered codebase intelligence — enabling developers and AI systems to ask questions about large codebases and get accurate, context-aware answers. The market timing for this investment is obvious in retrospect: code is one of the primary domains where LLMs are proving useful, and the ability to reason meaningfully about large existing codebases is one of the hardest problems in making LLM-based coding assistance genuinely reliable.
But the reason for the investment was not timing. It was the team's insight into why the problem is hard and why their approach addresses the hardness.
Why Codebase Intelligence Is Hard
The naive approach to asking questions about a codebase is to embed it and do semantic search. This works for simple queries — "find where this function is defined," "show me examples of this pattern." It fails for complex queries that require understanding cross-file dependencies, architectural relationships, implicit conventions, and the historical evolution of design decisions. A large production codebase is not a static collection of documents. It is a complex artifact with structure, history, and meaning that naive retrieval completely fails to capture.
The Greptile approach treats codebase understanding as a first-class engineering problem: building specialized indexing, retrieval, and context assembly pipelines that are designed specifically for code rather than adapted from document retrieval. The details of their approach — how they model dependencies, how they handle context assembly for multi-file questions, how they maintain index freshness — are where the real technical depth lives.
The API Thesis
What excited us specifically about Greptile, rather than other teams approaching similar problems, was the API-first product approach. Rather than building a developer tool that competes with GitHub Copilot, Greptile is building the underlying codebase intelligence layer that other AI coding tools can call into. An API that can answer "what does this function depend on across the codebase" or "explain the tradeoffs of changing this interface" is useful to code review agents, documentation generators, onboarding assistants, security analysis tools, and dozens of other products. The infrastructure bet is more durable than any specific application.