Budget-Friendly Code Review That Still Finds Real Bugs

It means a different trade, made deliberately. Budget Mode has lower bug detection and lower precision than Balanced, Precise, or Ultra — that is what makes it fast and inexpensive. But the baseline it trades down from is a strong one: Macroscope tops the Martian open-source code review benchmark as the most balanced code reviewer, and ranks #1 on F1 score. The point is that you no longer have to choose between a premium review and no review at all. Running Budget on routine changes and escalating the risky ones catches more bugs in total than reserving an expensive mode for the few PRs you can afford to spend it on.

Against CodeRabbit, the difference is depth and billing: on a benchmark of 118 real runtime bugs Macroscope detected roughly 48% versus CodeRabbit's 46%, using AST-based code walking for cross-file analysis rather than reviewing a diff in isolation, and Macroscope bills per KB reviewed while CodeRabbit charges per seat. Against Greptile, the difference is focus: Greptile centres on codebase Q&A and learning from PR comments, where Macroscope is built around cross-file bug detection plus explicit enforcement through Check Run Agents, again on usage-based rather than seat-based pricing. Full breakdowns are on the Macroscope vs CodeRabbit and Macroscope vs Greptile comparison pages.

Budget is $0.025 per KB, exactly half the $0.05 Balanced default, and one eighth the $0.20 Ultra rate. Because Detection Mode is set per review, the saving is not all-or-nothing: you can default a repository to Budget and still spend Ultra money on the pull requests that touch production-critical paths.

Yes, and that is the intended pattern. Set a repository default in Settings → Repos, pin a personal default in Settings → Personal, override a single pull request with a macroscope-budget or macroscope-ultra label, or ask for one review in a specific mode by mentioning @macroscope-app in a comment. A label override applies to every subsequent push on that PR.

It depends on whether they resolve to different modes. Labels are matched on the mode they resolve to, not on the literal string, so several labels naming the same mode — a different casing, or an accepted alias like macroscope-balanced and macroscope-recall — count as one override and the review runs normally. Only labels resolving to two or more distinct modes are a genuine conflict. In that case the correctness check is skipped for that pull request rather than run in a guessed mode, scoped to correctness only: the approvability check and any Check Run Agents you have enabled still run. To resolve it, remove the labels that name the other mode so the rest resolve to one, or name the mode explicitly by mentioning @macroscope-app in a comment, which makes that run unambiguous.

You start with $100 in free usage and 1,000 free agent credits every month, with no card required. At Budget Mode rates that $100 covers roughly 208 average-sized pull requests before you pay anything.

All of them. Detection Mode changes how much compute a review gets, not which languages Macroscope can read. Connect any repository and reviews run on every pull request automatically.

Under five minutes. Install the GitHub app, connect a repository, and set Detection Mode to Budget in Settings → Repos. There is nothing to install in your codebase and no configuration file to commit.