Tabnam

Tabnam, Inc. is the Delaware legal entity behind Greptile, not a currently marketed company called Tabnam. The history has two distinct product phases and founder/team lists often mix them. Georgia Tech computer-science students Daksh Gupta and Soohoon Choi began Tabnam in a capstone/CREATE-X program in 2022. The initial team also included Robert 'Davis' Liddell and Ethan Perry and built an AI shopping assistant that scraped online user-experience data; it later shifted to collecting conversational customer feedback over SMS/WhatsApp and analyzing it for consumer brands. Tabnam was an InVenture Prize finalist and raised an $800,000 pre-seed in 2023 from Drive Capital, consumer entrepreneur Tristan Walker and ISS founder Chris Klaus. After moving from Atlanta to San Francisco, the team was initially rejected by Y Combinator and questioned the feedback-market thesis. At a July 2023 Scale AI/SoMa hackathon, Gupta, Choi and Gupta's Georgia Tech roommate Vaishant Kameswaran wrote a graph-style codebase indexing system. They won paying customers within a week, reached roughly 100 users before joining Y Combinator W24 and pivoted the corporation to Greptile. Current first-party and YC sources consistently identify the Greptile co-founders as Gupta, Choi and Kameswaran; Liddell and Perry are important original Tabnam team members but are not presented as current Greptile founders. Gupta is co-founder/CEO and president; Choi is co-founder/CTO; Kameswaran is a co-founder/technical leader. All studied computer science at Georgia Tech, with Kameswaran also minoring in linguistics; Gupta is from Noida, India, Choi from South Korea and Gupta previously interned at Amazon. Greptile began as an API/chat interface for codebase-optimized retrieval-augmented generation. Instead of treating a repository as isolated files, it builds a graph of files, functions/classes and dependencies, follows references to retrieve context and lets LLMs answer questions, enrich Sentry alerts, update documentation, review merge requests and add context to Jira/Linear work. Product focus narrowed around automated pull-request review, then expanded into an independent validation layer for human- and AI-written code. The GitHub/GitLab app indexes the repository, reviews every PR beyond the diff, posts severity-rated inline findings, summaries, diagrams, confidence scores and suggested fixes, and can send findings to Claude Code, Codex, Cursor, Conductor or Devin. Custom rules and organization-isolated memories learn preferences from reactions, replies and human review comments. An MCP server, CLI, Claude plugin and integrations connect the review loop to coding agents and external work systems. July 2026 security review combines Opengrep/static patterns, software-composition/dependency checks and an AI exploit agent. Model inversion experimentally routes GPT-authored changes to Claude-family review and vice versa. TREX, introduced in June 2026 and still labeled beta, writes/chooses tests, starts services, mocks inputs, calls APIs and browser agents, runs the PR branch inside a sandbox and attaches logs, screenshots, traces, scripts or video; Greptile says execution adds roughly 20% more bugs than static/agent review alone. Version 5, enabled for all users August 5, 2026, uses parallel narrowly scoped agents, each exploring a bug hypothesis. A production A/B test over more than one million PRs reported median review time falling from 5:04 to 2:25, author-addressed comments rising from 52% to 66% and positive replies per PR increasing 28.6%. These are company-observed product metrics, and 'addressed' can be LLM-classified; they are not an independent randomized benchmark of escaped production defects. Funding comprises the $800,000 2023 pre-seed, a $4.1 million seed announced June 2024 led by Initialized Capital with Y Combinator and angels Rich Aberman and JJ Fliegelman, and a $25 million Series A announced September 2025 led by Benchmark with continued backing from Initialized, YC and Cory Levy. That sums to $29.9 million and Georgia Tech reports $30 million cumulative. Benchmark partner Eric Vishria joined the board. Georgia Tech reports the Series A valued Greptile at $180 million, without specifying pre- or post-money. The August 2025 Form D is frequently misread as a $41,413,374 financing: its clarification says exactly $24,999,999 was new cash and the remaining $16,413,375 represented conversion of existing convertible securities and secondary exchanges. The filing therefore describes Series A recapitalization mechanics and is not $41.4 million of additional funding. It names 27 investors/purchasers and Gupta, Choi and Vishria as related persons. As of January 2026 Georgia Tech reported more than 2,000 customers; by June-August Greptile claimed over 22,000 teams. The Series A release reported 500 million lines reviewed and 180,000+ bugs prevented in one month for users including Brex, Substack, PostHog, Bilt and YC's internal software team. The current site lists customers/logos including NVIDIA, Klaviyo, Retool, Scale AI, Mintlify, Zapier, Joby, WorkOS, Compass, Crossmint, Browserbase and Material Security. These are vendor claims and do not disclose contract value, active paid seats or retention. YC listed 35 people in August 2026, up from roughly a dozen in mid-2025. Current pricing is product-led plus enterprise: Starter is free for one active developer and 50 monthly credits; Pro is $30 per seat per month including 50 credits per seat, with additional credits at $1; a standard review uses one credit and a TREX review three; Enterprise is custom-priced with self-hosting, SSO/SAML, GitHub Enterprise, dedicated support, custom DPA/invoicing and terms. Annual/multi-year discounts, free qualified MIT/Apache open-source use and 50% early-startup discounts exist. Enterprise/API economics, GPU/LLM gross margin, ARR, NRR, customer concentration and profitability are undisclosed. The company supports GitHub/GitLab, cloud and Docker/Kubernetes/air-gapped self-hosting. Greptile says cloud systems use AWS and Azure plus OpenAI and Anthropic APIs, code is cached on encrypted filesystems until GitHub/GitLab access is revoked, vector embeddings of paths/docs/generated docstrings are stored, and chat logs may be accessed for support unless logging is disabled. It may use aggregated/anonymized customer data for analytics and AI training; customers can opt out. Self-hosted customers can keep code/logs in their own environment and bring/self-host LLMs. It claims SOC 2 Type II compliance and deletion controls, with enterprise reports/DPA available on request. These controls matter because Greptile receives extremely sensitive source code, history, comments, dependencies and potentially secrets; a breach, cross-tenant retrieval error, subprocessor misuse, insider access or imperfect de-identification could expose customer IP or vulnerabilities. TREX adds untrusted-code execution and sandbox-escape, network/secret-exfiltration and supply-chain risks. Product errors also matter: stochastic reviewers can hallucinate, miss defects, produce noisy comments or create automation bias; performance claims do not prove causation or coverage across languages/repositories. Customers must retain human review, tests and security controls. Competition is intense from CodeRabbit, Graphite Diamond, Cursor Bugbot, GitHub Copilot code review, Qodo, Bito, Sourcegraph, Snyk/Sonar/semgrep-style tools and internal agents. CodeRabbit's August 2026 $1.5 billion round and large customer/review scale show both market validation and pressure. Greptile differentiates around repository-wide graph context, independence from the code-generating model, organization learning, self-hosting and runtime evidence. A material reputational/employment controversy began in November 2024 when Gupta publicly described typical 9 a.m.-11 p.m. workdays, Saturdays and no work-life balance, saying candidates were told up front. The post drew widespread criticism and reported threats; Gupta later continued advocating a 9-9-6 culture. This is not a legal finding, but creates recruiting, retention, burnout, discrimination/accommodation, wage-hour and key-person risk. No material Greptile-specific lawsuit, regulator action, breach or outage-related proceeding was found in manual review. The highest-priority diligence gaps are audited ARR/growth/retention and inference margins; active paid customers versus free teams; independent false-negative/false-positive and production-defect outcomes; exact data flows/subprocessors/retention and opt-out defaults; TREX isolation; employee attrition; cap table/secondary mechanics; and durability against model/platform incumbents.

Founding story

A Georgia Tech project called Tabnam moved from AI shopping/user-experience search to conversational customer feedback. After relocating and questioning that market, Gupta, Choi and Kameswaran built a codebase graph index at a July 2023 San Francisco hackathon, won paying customers within a week and pivoted to Greptile before YC W24.

Business model

B2B developer-tools SaaS and usage platform for automated code review, security analysis and runtime validation, sold self-serve per developer/credit and through custom enterprise cloud or self-hosted deployments.

Free-to-paid product-led conversion; $30/developer/month Pro subscription including review credits, $1 overage credits, and custom enterprise/self-host/API annual or multi-year contracts. Inference, indexing, storage and support are major cost drivers.

Traction

Nearly 400 teams at the Jun 2024 seed; 500m lines and 180k+ bugs claimed in Sep 2025 month; 2,000+ customers by Jan 2026; 22,000+ teams by Jun-Aug 2026; 35 YC-listed people. V5 production test spanned 1m+ PRs. Active paid seats, ARR, retention and independent defect outcomes undisclosed.

Latest developments

Greptile launched v4 and credit pricing in March 2026, TREX runtime validation in June, default security review and model inversion in July, and v5's parallel bug-hypothesis agents on August 5. Current site claims 22,000+ teams; v5 production data reports faster reviews and higher address/positive-reply rates.