Greptile
Last reviewed
Jun 4, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 · 2,094 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 4, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 · 2,094 words
Add missing citations, update stale details, or suggest a clearer explanation.
Greptile is an AI code-review tool that indexes an entire codebase and uses that full-repository context to review pull requests, flagging bugs, regressions, and style violations that require cross-file understanding. The product is built by Greptile, Inc. (greptile.com), a San Francisco company founded in 2023 by Georgia Tech graduates Daksh Gupta, Soohoon Choi, and Vaishant Kameswaran. Greptile went through the Y Combinator Winter 2024 batch and is backed by a $4.1 million seed round and a $25 million Series A led by Benchmark. Its core idea is that a large language model reviewing a code change does a far better job when it can see how that change interacts with the rest of the repository, so Greptile builds and continuously updates a searchable index of the whole codebase before it comments on any pull request.
Greptile is one of a wave of AI code-review startups, alongside CodeRabbit, Graphite, and Diamond, that emerged as generative AI made automated, context-aware review of pull requests practical. It competes more broadly in the AI developer-tools market with ai code generation products such as GitHub Copilot, Cursor, and Windsurf, though Greptile focuses on review rather than authorship.
Greptile's founders, Daksh Gupta (CS 2023), Soohoon Choi (CS 2023, Math 2023), and Vaishant Kameswaran, met at the Georgia Institute of Technology and went through the university's CREATE-X startup-launch program. Their first project was an AI shopping assistant called Tabnam; the team pivoted after customer discovery suggested more demand in software-development tooling. (The "tabnam" name still survives as the company's Crunchbase organization slug.) Gupta has credited CREATE-X with introducing him to Choi and giving the team the confidence to treat starting a company as a career path. Early on, the founders raised a $100,000 investment from Georgia Tech alumnus and security entrepreneur Christopher Klaus, which helped them relocate to San Francisco.
The product that became Greptile began as a July 2023 hackathon project built around the simple idea of letting developers "talk" to their codebase, and the company initially operated under the name Onboard (also referred to as Onboard AI), positioning the tool as a way to help engineers onboard onto unfamiliar code. Over roughly seven months, as thousands of developers used it, the founders concluded the tool was useful to all engineers slowed down by large codebases, not just new hires, and renamed it Greptile. The name is a play on grep, the Unix command-line utility for searching text across files. The company was accepted into Y Combinator's Winter 2024 batch, where it relaunched publicly as Greptile.
Greptile announced a $4.1 million seed round on June 9, 2024, led by Initialized Capital with participation from Y Combinator and a group of angel investors. At the time, the company reported moving from roughly 100 early users paying $10 to $20 per month to about 500 paying customers.
In July 2025, TechCrunch reported that Benchmark was in talks to lead a roughly $30 million Series A that would value Greptile at about $180 million, with Benchmark general partner Eric Vishria leading the deal. Greptile formally announced the round on September 23, 2025, describing it as a $25 million Series A led by Benchmark Capital, with continued participation from Y Combinator, Initialized Capital, and angel investor Cory Levy. Counting the seed, the company has raised on the order of $30 million in total. Georgia Tech, in a January 2026 profile, cited the $180 million valuation and the $30 million total figure.
| Round | Date | Amount | Lead investor | Notes |
|---|---|---|---|---|
| Seed | June 9, 2024 | $4.1M | Initialized Capital | With Y Combinator and angels |
| Series A | September 23, 2025 | $25M | Benchmark (Eric Vishria) | ~$180M valuation; ~$30M raised to date |
By early 2026, Georgia Tech reported Greptile had more than 2,000 customers, including Brex, Whoop, and Substack.
Greptile's central technical claim is that it gives an LLM "complete context" of a repository before review. Rather than embedding raw source code, the founders described an indexing pipeline that parses the abstract syntax tree (AST) of a codebase using tree-sitter, recursively generates natural-language docstrings for each node in the tree, and then embeds those docstrings. Retrieval combines vector search and keyword search with an agentic step: an agent evaluates the relevance of search results and follows references through the source code to pull in related context. This index is maintained in real time and updated as the code changes, so reviews reflect the current state of the repository. The approach is a specialized form of retrieval-augmented generation tailored to large codebases, and the founders have argued it works better than naive RAG over source files.
The codebase index underpins three kinds of use: querying a repository in natural language (for example, "How does authentication work in this code base?"), an API for building custom tools such as PR-review bots, and Greptile's own automated code-review product. Gupta has said Greptile deliberately uses state-of-the-art models, chained across multiple steps, and feeds them as much codebase context as possible, which he has cited as the reason the product is priced at a premium to rivals.
When connected to a GitHub or GitLab repository, Greptile posts review comments on pull requests. Because it can search the whole codebase, it aims to catch issues that depend on cross-file understanding, such as a change that breaks a caller elsewhere, rather than only commenting on the diff in isolation. The company describes the bot as functioning like an experienced coworker with deep knowledge of the customer's code. Greptile supports custom rules, learns a team's conventions from past review comments, and can self-host inside a customer's own cloud, including an air-gapped virtual private cloud, for security-sensitive organizations. It advertises SOC 2 Type II compliance and SSO/SAML on its enterprise tier.
Greptile has rebuilt its product several times as it grew:
| Version | Date | Summary |
|---|---|---|
| Greptile 2.0 | October 8, 2024 | Ground-up rewrite into an extensible platform: code-review bot, Jira/Linear ticket generation, custom integrations (Sentry, Datadog, test pipelines, docs) |
| Greptile v3 | September 23, 2025 (detailed Nov 26, 2025) | Rewrite of the core agent into an agentic loop that recursively searches the codebase for multi-hop reasoning; adds rule "learning" from past comments, an MCP server, Jira/Notion integrations, and auto-detection of CLAUDE.md and .cursorrules files |
| Greptile v4 | March 5, 2026 | Updated review agent and a base-plus-usage pricing model |
With v3, Greptile moved from a fixed flowchart of steps (read the diff, gather codebase context, gather external context, write comments) to an agentic loop in which the system runs with tools such as codebase search and access to learned rules, letting it follow nested function calls and do what the company calls multi-hop thinking. The v3 release added a Model Context Protocol (MCP) server so that coding AI agents and IDEs can read Greptile's comments and resolve them automatically. The company reported that v3 reviewed more than one billion lines of code after launch and improved its comment acceptance ("action") rate substantially over v2. The later v4 release reported, from a month-long A/B test across customers, increases such as addressed comments per pull request rising from 0.92 to 1.60.
Greptile publishes a free tier for qualifying non-commercial open-source projects (MIT, Apache, or GPL licensed), a paid Pro plan, and a custom Enterprise plan.
| Plan | Price | Includes |
|---|---|---|
| Open source | Free | Qualified non-commercial projects (MIT/Apache/GPL) |
| Pro | $30 per developer / month | 50 reviews per seat per month, then $1 per additional review; unlimited repos, users, and custom rules |
| Enterprise | Custom | Self-hosting, SSO/SAML, GitHub Enterprise, dedicated support, custom DPA |
Greptile offers a 14-day free trial and a 50% discount for early-stage startups (pre-Series A with under $2 million in trailing-12-month revenue). With the v4 release in March 2026, the company moved to the "base plus usage" model shown above, replacing a flat per-seat subscription; it framed the change as a response to AI coding agents sharply increasing the number of commits being reviewed, and said fewer than 10% of active users would exceed the included 50 reviews. Greptile's API has been billed per "unit," with a standard query counting as one unit and a more powerful "genius" query counting as three.
Greptile has reported rapid usage growth. The company said that in a single month it reviewed more than 500 million lines of code, prevented more than 180,000 bugs, and processed over 700,000 pull requests. Customers it has named publicly include Brex, Substack, PostHog, Bilt, Whoop, Raycast, Podium, and Vouch, and the company has said Y Combinator's internal software team is a user. Greptile and third-party profiles describe it serving from several hundred to a few thousand software teams across more than 50 countries as it scaled from 2024 into 2026.
Greptile and Gupta drew broad media attention in late 2024 for public statements about an extreme work culture. In November 2024, Gupta posted on X that he had begun telling job candidates in the first interview that Greptile "offers no work-life-balance," that typical workdays "start at 9am and end at 11pm, often later," and that the team works Saturdays and sometimes Sundays. He framed the disclosure as a way to let people who do not want that environment opt out. The post was widely shared, and the episode was covered by outlets including KRON4 and NBC Bay Area.
A second, larger wave followed a December 16, 2024 San Francisco Standard article on intense startup culture, in which Gupta was quoted describing the company's ethos as: "The current vibe is no drinking, no drugs, 9-9-6, lift heavy, run far, marry early, track sleep, eat steak and eggs." "9-9-6" refers to working from 9 a.m. to 9 p.m., six days a week, a schedule associated with parts of China's tech industry. The quote spread well beyond the original article and made Gupta a focal point in a wider debate about long working hours in technology.
In a September 10, 2025 blog post titled "Greptile's work culture," the company sought to clarify its position. Gupta wrote that he does not mandate specific weekly hours and does not tell people to work hard; instead, Greptile selects candidates who want a high-velocity, early-stage environment, offers generous equity, and sets the pace by founder example. He argued that software tends to be winner-take-all, so a startup that does not push as hard as possible "may as well not try," while also saying such intensity should last only a year or two rather than indefinitely. Gupta has said the original viral posts were not a deliberate marketing effort, though commentators noted the attention raised Greptile's profile.