Coding agent can create huge code diffs. Someone still has to read it. seendiff is a local diff tool for large changes. Track what you've seen, review in chunks, and pull the coding agent in to explain its own work.
Loading comments…
Maker
kettle_dev
Visit Website
seendiff.com
Project Info
Product Keywords
seendiff is a local diff review tool built for the reality of AI-assisted development: coding agents can generate enormous diffs in minutes, but someone still has to read them. It turns that chore into a structured, trackable process. Running as a localhost server inside any git repository, seendiff lets you mark code blocks as seen or reviewed, navigate changes with keyboard shortcuts, and even pull your coding agent back in to explain its own work. It's a Python CLI that needs Python 3.10+ and git, installs in seconds, and keeps all your review progress local.
Code blocks are marked seen automatically when they scroll through your viewport, while reviewed is an explicit action — press R for a block or f for the whole file. This two-tier system means you always know what you've merely glanced at versus what you've actually verified.
Everything is accessible from the keyboard: ←/→ moves between change blocks, ↑/↓ between files, n/N jumps to the next unreviewed block across files, and V marks all seen-but-unreviewed blocks in a file as reviewed. No mouse needed.
Nothing leaves your machine — no accounts, no telemetry, no network calls outside of localhost. The tool has minimal dependencies (just four), installs in seconds, and requires no configuration file.
Designed with scale in mind, seendiff handles 40,000-line files smoothly thanks to virtualization. A minimap down the side of the file shows at a glance what's seen, unseen, and unreviewed, and search covers every changed file at once.
Other tools you might consider
The coding agent wrote the diff — seendiff makes it explain itself.
That's the real edge: seendiff isn't just a viewer, it's a bridge back to the agent. The bundled skills/seendiff skill lets your coding agent open a diff and write a walkthrough of its own work, which you can then review against the actual changes. Combined with the seen/reviewed tracking that survives rebases and amends, this turns chaotic AI-generated changes into a structured, auditable review process.
You're tired of eyeballing massive diffs from coding agents without any system, or you want a fast, private, keyboard-driven review tool that doesn't phone home. If you value your review progress persisting across git operations and want the ability to script agent walkthroughs, seendiff is worth a look. It's especially useful if you're already using Claude Code or similar tools and want a tighter feedback loop between generating changes and verifying them.
