Ellipsis (software)
Last reviewed
Jun 4, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 · 1,837 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 · 1,837 words
Add missing citations, update stale details, or suggest a clearer explanation.
Ellipsis (stylized in lowercase as ellipsis, at ellipsis.dev) is an AI-powered code-review tool that automatically reviews GitHub pull requests and can make code changes in response to comments. The product installs as a GitHub app: when a pull request is opened, Ellipsis uses large language model agents to flag logical bugs, anti-patterns, security issues, and style-guide violations, and developers can tag the bot (@ellipsis-dev) in a comment to have it write a fix, answer a question about the codebase, or generate new code. Ellipsis was founded in 2023 by Hunter Brooks and Nick Bradford and took part in Y Combinator's Winter 2024 (W24) batch. The company raised a $2 million seed round in April 2024 and positions itself in the crowded market for AI code review alongside tools such as CodeRabbit, Greptile, and Graphite Reviewer.
Ellipsis was founded in 2023 by Hunter Brooks (co-founder and chief executive) and Nick Bradford (co-founder). On Crunchbase the company is recorded under the legal/former name BitBuilder. The two founders met through backgrounds in machine learning and software engineering. According to the company's Y Combinator launch profile, Brooks had previously worked as a machine-learning engineer at Amazon Web Services and Amazon Ads, with earlier roles at the fintech firms Brigit and Capital One, and had published astrophysics research before moving into programming. Bradford had been a founding engineer at Grit (an AI codebase-migration startup), had done LLM interpretability research at Redwood Research, had managed a team at Hyperscience, and had built trading algorithms at Bridgewater Associates. The startup is based in New York City.
Ellipsis was accepted into Y Combinator's Winter 2024 batch. It made a public Show HN launch on Hacker News on February 28, 2024, under the title "Show HN: Ellipsis, Automatic pull request reviews," describing a tool that reviews a pull request automatically whenever one is opened, framed as a step toward an "AI software engineer." A second Show HN, "Ellipsis, Automated PR reviews and bug fixes," followed in May 2024.
In 2025, Nick Bradford's public profiles indicated that he had moved on to work at the AI coding company Cursor on cloud agents; the Y Combinator company page continued to list both Brooks and Bradford as co-founders, with Brooks as the active CEO.
Public records on Crunchbase show two early financing events, followed by the announced seed round.
| Round | Date | Amount | Notes |
|---|---|---|---|
| Pre-seed | February 27, 2024 | Undisclosed | Recorded on Crunchbase (under the BitBuilder entity) |
| Seed | April 11, 2024 | $2 million | Announced publicly by the company |
The company announced the $2 million seed round on its blog. Named participants included the venture firms Twenty Two Ventures, Four Cities Capital, Garage Capital, Amino Capital, Transpose Platform Management, and Pioneer Fund, along with a group of Y Combinator partners and angels including Dalton Caldwell, David Lieb, Garry Tan, Brad Flora, and Gustaf Alstromer. Ellipsis was later included in Business Insider's October 2024 roundup of AI coding startups that had raised the most venture funding.
Ellipsis is a cloud-based GitHub application rather than a local tool or IDE plugin. Its workflow is built around the pull request:
@ellipsis-dev (for example, "fix this") and Ellipsis will generate a patch. The company says it returns "working, tested code," and changes are proposed as separate pull requests that a human must approve before merging.The company reports integrations with GitHub (required), Slack, and Linear. As of 2026 its documentation and third-party reviews describe the product as GitHub-only, with no support for GitLab, Bitbucket, or Azure DevOps, and no self-hosted option.
Co-founder Nick Bradford has described Ellipsis's architecture in technical write-ups. Rather than a single large agent driven by one big prompt, Ellipsis is built from many smaller LLM agents that can be independently benchmarked and optimized, an approach the company calls decomposition. This lets the system mix different underlying models for different tasks; Bradford has cited using both OpenAI's GPT-4o and Anthropic's Claude (Sonnet 3.5), and has noted experimenting with OpenAI's o1 for harder problems.
The review pipeline runs multiple "comment generators" in parallel, each looking for a different class of issue (such as custom-rule violations or code duplication). Their output passes through filtering stages that deduplicate overlapping comments, check for hallucinations using attached evidence links, apply a confidence threshold that customers can configure, and incorporate user feedback through an embedding search over previously rated comments. A separate code-search subagent uses multi-step retrieval-augmented generation combining keyword and vector search, with tree-sitter abstract-syntax-tree parsing to chunk code at the function and class level and LLM-generated file summaries for higher-level queries. Instead of relying on a simple similarity cutoff, an LLM-based classifier selects which code is relevant given the agent's current context.
Operationally, GitHub webhook events are routed (via Hookdeck) to a FastAPI web application and then onto a Hatchet workflow queue, an asynchronous design that the company says prioritizes accuracy over latency. For evaluation, the team builds small benchmark sets, uses LLM-as-judge grading (made more reliable for code search by requiring verifiable evidence), and runs an "agent trajectory auditor" that diagnoses where an agent failed. Bradford has said the team favors prompt iteration, few-shot examples, and composability over fine-tuning.
Ellipsis states that it does not persist customer source code: code exists only in memory on its servers, inside a private AWS virtual private cloud, while a job is being processed, and is not stored afterward. The company says it holds a SOC 2 Type I certification.
Ellipsis publishes a simple pricing model. Public GitHub repositories are reviewed for free with no feature limits, which the company markets to open-source projects. For private repositories, Ellipsis charges a flat per-developer fee for unlimited use, reported at $20 per developer per month (some third-party comparisons cite $19), with a 7-day free trial.
Ellipsis received early attention within the Y Combinator and startup founder community for the quality of its reviews. The company's headline claim is that teams using Ellipsis merge pull requests about 13% faster on average than teams without AI code review. To measure work, Ellipsis says it uses a "Units of Work" metric based on change complexity rather than raw lines of code.
The company's self-reported adoption has grown over time. At its February 2024 launch it cited a small number of early customers; by the time of the seed announcement it reported over 90 companies using the product, and by 2026 its marketing cited figures in the range of hundreds of companies and tens of thousands of GitHub repositories, with named customers including the terminal company Warp and the LLMOps startup PromptLayer. These usage figures are company-stated and have not been independently audited.
In independent comparisons of AI code-review tools published in 2025 and 2026, reviewers generally placed Ellipsis among the credible options while noting trade-offs. Coverage praised its low-noise reviews and its ability to apply fixes on command, and in at least one hands-on test a reviewer reported that Ellipsis caught a React state-management bug that other tools and human reviewers had missed. Common criticisms were that it supports only GitHub, that it is weaker on less common languages such as Go and Rust, and that it is less suited to deep architectural analysis.
Ellipsis competes in a fast-growing category of AI code-review and pull-request automation products. Its closest comparisons in independent reviews are:
| Tool | Notable contrast with Ellipsis |
|---|---|
| CodeRabbit | Supports more platforms (GitHub, GitLab, Bitbucket, Azure DevOps) and tends to generate more comments; widely treated as the mature default. |
| Greptile | Builds a retrieval index over the whole codebase for cross-file, context-aware reviews; typically priced higher. |
| Graphite Reviewer | Part of Graphite's stacked-pull-request workflow and code-review platform. |
| Qodo (formerly Codium) | Another AI review and test-generation tool frequently benchmarked alongside Ellipsis. |
Ellipsis differentiates itself less on breadth of platform support and more on combining review with action: it not only comments on a pull request but can also write the fix, answer questions, and generate code from a comment, reflecting the founders' stated long-term goal of an autonomous AI software engineer. The broader space is also influenced by general-purpose AI coding tools such as Cursor and GitHub Copilot, the latter of which has added its own pull-request review feature.
CodeRabbit - Greptile - Cursor - GitHub Copilot - AI code generation - AI agents - Large language model - Retrieval-augmented generation - Y Combinator