Here’s a demo video: https://www.youtube.com/watch?v=pglEoiv0BgY
We (Allis and Paul) are engineers who faced this problem when we worked together at our last startup. Code review quickly became our biggest bottleneck—especially as we started using AI to code more. We had more PRs to review, subtle AI-written bugs slipped through unnoticed, and we (humans) increasingly found ourselves rubber-stamping PRs without deeply understanding the changes.
We’re building mrge to help solve that. Here’s how it works:
1. Connect your GitHub repo via our Github app in two clicks (and optionally download our desktop app). Gitlab support is on the roadmap!
2. AI Review: When you open a PR, our AI reviews your changes directly in an ephemeral and secure container. It has context into not just that PR, but your whole codebase, so it can pick up patterns and leave comments directly on changed lines. Once the review is done, the sandbox is torn down and your code deleted – we don’t store it for obvious reasons.
3. Human-friendly review workflow: Jump into our web app (it’s like Linear but for PRs). Changes are grouped logically (not alphabetically), with important diffs highlighted, visualized, and ready for faster human review.
The AI reviewer works a bit like Cursor in the sense that it navigates your codebase using the same tools a developer would—like jumping to definitions or grepping through code.
But a big challenge was that, unlike Cursor, mrge doesn’t run in your local IDE or editor. We had to recreate something similar entirely in the cloud.
Whenever you open a PR, mrge clones your repository and checks out your branch in a secure and isolated temporary sandbox. We provision this sandbox with shell access and a Language Server Protocol (LSP) server. The AI reviewer then reviews your code, navigating the codebase exactly as a human reviewer would—using shell commands and common editor features like "go to definition" or "find references". When the review finishes, we immediately tear down the sandbox and delete the code—we don’t want to permanently store it for obvious reasons.
We know cloud-based review isn't for everyone, especially if security or compliance requires local deployments. But a cloud approach lets us run SOTA AI models without local GPU setups, and provide a consistent, single AI review per PR for an entire team.
The platform itself focuses entirely on making human code reviews easier. A big inspiration came from productivity-focused apps like Linear or Superhuman, products that show just how much thoughtful design can impact everyday workflows. We wanted to bring that same feeling into code review.
That’s one reason we built a desktop app. It allowed us to deliver a more polished experience, complete with keyboard shortcuts and a snappy interface.
Beyond performance, the main thing we care about is making it easier for humans to read and understand code. For example, traditional review tools sort changed files alphabetically—which forces reviewers to figure out the order in which they should review changes. In mrge, files are automatically grouped and ordered based on logical connections, letting reviewers immediately jump in.
We think the future of coding isn’t about AI replacing humans—it’s about giving us better tools to quickly understand high-level changes, abstracting more and more of the code itself. As code volume continues to increase, this shift is going to become increasingly important.
You can sign up now (https://www.mrge.io/home). mrge is currently free while we're still early. Our plan for later is to charge closed-source projects on a per-seat basis, and to continue giving mrge away for free to open source ones.
We’re very actively building and would love your honest feedback!
Automated review tools like this are especially important for an open source project because you have to maintain a quality bar to keep yourself sane but if you're too picky then no one from the community will want to contribute. AI tools are like linters and have no feelings, so they will give the feedback that you as a reviewer may have been hesitant to give, and that's awesome.
Oh, and on the product itself, I think it's super cool that it comes up with rules on its own to check for based on conventions and patterns that you've enforced over time. E.g. we use it to make sure that all function calls that pull from an upstream API are decorated with our standard error handler.
It would be awesome if the custom rules were generalized on the fly from ongoing reviewer conversations. Imaging two devs quibble about line length in a PR, and in a future PR, the AI reminds about this convention.
Would this work seamlessly with AI Engineers like Devin? I imagine so.
This will be very handy for solo devs as well, even those who don't use Coding CoPilots could benefit from an AI reviewer, if it does not waste their time.
Maybe there can be multiple AI models review the PR at the same time, and over time, we promote the ones whose feedback is accepted more.
On working with Devin: Yes, right now we're focused on code review, so whatever AI IDE you use would work. In fact, it might even be better with autonomous tools like Devin since we focus on helping you (as a human) understand the code they've written faster.
Interesting idea on multiple AI models --we were also separately toying with the idea of having different personas (security, code architecture), will keep this one in mind!
And I absolutely love your idea of having multiple AI models review PRs simultaneously. Benchmarking LLMs can be notoriously tricky, so a "wisdom of the crowds" approach across a large user base could genuinely help identify which models perform best for specific codebases or even languages. We could even imagine certain models emerging as specialists for particular types of issues.
Really appreciate these suggestions!
(mixture of 400 lines of C and 100 lines of Python)
It also didn't flag the one SNAFU that really broke things (which to be fair wasn't caught by human review either, it showed in an ASAN fault in tests)
We take all these into consideration when improving our AI, and your direct reply will fine tune comments for your repository-only.
(We already have problems with our human review being too superficial; we've recently come to a consensus that we're letting too much technical debt slip in, in the sense of unnoticed design problems.)
Now the funny part is that I'm talking about a FOSS project with nVidia involvement ;D
But also: this being a FOSS project, people have opened AI-generated PRs. Poor AI-generated PRs. This is indirectly hurting the prospects of your product (by reputation). Might I suggest adding an AI generated PR detector, if possible? (It's not in our guidelines yet but I expect we'll be prohibiting AI generated contributions soon.)
if you have specific feedback on the pr--feel free to email at contact@mrge.io and i'll take a look personally and see if we can adjust anything for your repo.
nice idea on the fully AI-generated PRs! something in our roadmap is to better highlight PRs or chunks that were likely auto-gened. stay tuned !
AI or conventional bots for PRs are neat though. Where I work we have loads of them checking all sorts of criteria. Most are rules based. E.g. someone from this list must review if this folder changes. Kinda annoying when getting the PR in but overall great for quality control. We are using an LLM AI for commenting on potential issues too. (Sorry I don't have any influence to help them to consider yours)
If that's it, we actually support stacked PRs (currently in beta, via CLI and native integrations). My co-founder, Allis, used stacked PRs extensively at her previous company and loved it, so we've built it into our workflow too. It's definitely early-stage, but already quite useful.
Docs if you're curious: https://docs.mrge.io/overview
In the meantime, good luck with that hairy review—hope it goes smoothly! If you're open to it, I'd love to reach out directly once GitLab support is ready.
Appreciate the feedback around security as well; protecting against supply-chain attacks is definitely top of mind for us as we build this out.
I don't remember adding that feature so it might be a bug
After watching it sit there for 5-10 seconds before loading the section 'data-framer-name="Join"' I decided to inspect the element after it did load to see what it was doing. That's when I spotted all the JS and data attributes implying it was likely built with one of those drag-and-drop site builders, which explains why it may be behaving in an unexpected way for you. It also explains why it may default to "fade in on scroll" behavior, if my experience is any indication, because marketing folks _love_ that shit
The good news with mrge is that it works just like any other AI code reviewer out there (CodeRabbit, Copilot for PRs, etc.). All AI-generated review comments sync directly back to GitHub, and interacting with the platform itself is entirely optional. In fact, several people in this thread mentioned they switched from Copilot or CodeRabbit because they found mrge's reviews more accurate.
If you prefer, you never need to leave GitHub at all.
We would be happy to try except when it has write/merge permissions .
One click and auto merge are nice to have. Having the bot (and your company) able to deploy any code changes to production (by accident, via hack, etc) is a no go.
Suggest making them optional features and just having code comments/repo read version.
Not sure if it’s possible - but if the permissions could exclude specific branches that would be ok as well.
But needs to be no way a malicious actor could write/merge to main.
I watched your demo vid and the two things that stuck out to me were the summarizing of changes, grouping of file changes by concept, and the diagram generation. Graphite does generate summaries of PRs if you ask it to, but it's an extra step that replaces the user authored PR description. I see that you have stacked diff support too.
I probably don't want to spend the time/energy to migrate my team off Graphite anytime soon, but would be interested in evaluating mrge. Is the billing per reviewer of PRs or by author of PRs? And how long is the free trial? I'm always reluctant to sign up for limited time free trials because I don't know if I'll actually have time to commit to assessing the tool in that time window.
I'm on Bitbucket so will have to wait :)
And totally hear you on Bitbucket—it's definitely on our roadmap. Would love to loop back with you once we get closer on that front!
If the repo is several GB, will you clone the whole thing for every review?
for custom rules, we do handle large monorepos by allowing you to add an allowlist (or exclude list) via glob patterns.
Clicking the get started button immediately wants me to sign up with github.
Could you explain on the pricing page (or just to me) what the 'free' is? I'm assuming a trial of 1 month or 1 PR?
I'm somewhat hesitant to add any AI tooling to my workflows, however this is one of the use cases that makes sense to me. I'm definitely interested in trying it out, I just think its odd that this isn't explained anywhere I could find.
We'll try to make this clearer!
Did some self-research on Reddit about why (https://www.reddit.com/r/github/comments/1gtxqy6/comment/lxv...)
I feel like that’s being overlooked here a bit too briefly. Is your target market not primarily larger teams who are most likely to have some security and privacy concerns?
I guess is there something on the roadmap to maybe offer something later ?
If that's something your team might need, I'd love to chat more and keep you posted as we explore this!
So, buy this AI tool (Merge) to review code written by another AI tool?
Instead of a Code review tool, why not have it instead as a static analyzer? Overall, the whole process will take much less time.
Isn't cursor already the "cursor for code review?"
That said, that doesn't sound like something very useful when I already use an ai code editor for code review. And github already supports automations for ci/ci for ai tools for code review. Maybe I just don't see value in an extra tool for this.
If you're interested in Stack PRs, you should definitely check them out on Mrge. By the way, we natively support them (in beta atm): https://docs.mrge.io/ai-review/overview
Hope this is the right time, as this would be a huge time-saver for me
also, i tried some other ai review tools before. one big issue was always that they are too nice and even miss obvious bad changes. did you encounter these problems? did you mitigate this via prompting techniques or finetuning?
For applying code changes with one-click: we keep suggestions deliberately conservative (usually obvious one-line fixes like typos) precisely to minimize risks of breaking things. Of course, you should confirm suggestions first.
Regarding AI reviewers being "too nice" and missing obvious mistakes—yes, that's a common issue and not easy to solve! We've approached it partly via prompt-tuning, and partly by equipping the AI with additional tools to better spot genuine mistakes without nitpicking unnecessarily. Lastly, we've added functionality allowing human reviewers to give immediate feedback directly to the AI—so it can continuously learn to pay attention to what's important to your team.
How different it is from that?
We've heard from users who've tried both that our AI reviewer tends to catch more meaningful issues with less noise, that's really something you should try for yourself and find out! (The great thing is that it's really easy to start using)
Beyond the AI agent itself (which is somewhat similar to CodeRabbit), our biggest differentiation comes from the human review experience we've built. Our goal was to create a Linear-like review workflow designed to help human reviewers understand and merge code faster.
I'm not even going to add to this.
We've heard from users who've tried both that our AI reviewer tends to catch more meaningful issues with less noise, that's really something you should try for yourself and find out! (The great thing is that it's really easy to start using)
Beyond the AI agent itself (which is somewhat similar to Copilot), our biggest differentiation comes from the human review experience we've built. Our goal was to create a Linear-like review workflow designed to help human reviewers understand and merge code faster.