seendiff

Product Introduction

1. Definition: seendiff is a local, point-in-time diff tool and code review application designed specifically for analyzing large-scale code changes, such as those generated by AI coding agents (LLMs) and AI-powered development harnesses.
2. Core Value Proposition: It exists to solve the critical bottleneck of human-in-the-loop code review for AI-generated code. seendiff enables developers to systematically understand, track, and review massive diffs by providing a visual, stateful interface that persists review progress locally, ensuring no change is missed.

Main Features

1. Automated & Manual Review State Tracking: The tool automatically marks code blocks as "Seen" when they scroll into the viewport. Users can then explicitly mark blocks or entire files as "Reviewed." This dual-state system (Seen/Reviewed) is visualized in a minimap, providing an immediate, high-level overview of review progress across thousands of lines.
2. Local-First, Secure Architecture: seendiff runs entirely on the local machine. It starts a localhost server, requires no accounts, collects no telemetry, and makes no external network calls. All review state is stored locally and survives operations like rebase, amend, or closing the application.
3. Integrated AI Walkthrough Skill: It includes a bundled "skill" for AI coding agents (like Claude Code). Once installed, the agent can open the seendiff interface and generate a step-by-step walkthrough of its own code changes directly within the tool, facilitating a collaborative "AI explains to human" review workflow.
4. Performance-Optimized for Large Diffs: Built with scalability in mind, it uses virtualization techniques to handle files with up to 40,000 lines while maintaining smooth scrolling and responsive UI. It has minimal dependencies (four) for quick installation.
5. Fully Keyboard-Driven Navigation: Designed for efficiency, it offers comprehensive keyboard shortcuts for navigating between files (↑, ↓) and change blocks (←, →, n, N), marking review status (R, f, V), toggling views (s), and searching across all changed files (Ctrl+F).

Problems Solved

1. Pain Point: The overwhelming volume and complexity of code diffs produced by AI coding assistants make manual review tedious, error-prone, and difficult to track, leading to potential bugs and security issues slipping into the codebase.
2. Target Audience: Software engineers, tech leads, and open-source maintainers who regularly review large pull requests or merge requests, especially those working with AI pair programmers (e.g., GitHub Copilot, Cursor, Claude Code) or overseeing automated code generation.
3. Use Cases: Reviewing a multi-file refactor generated by an LLM; auditing the changes from a major dependency update; conducting a final sign-off on a feature branch with hundreds of changed lines; onboarding a new team member by having an AI walk them through a recent subsystem change.

Unique Advantages

1. Differentiation: Unlike traditional diff tools (e.g., git diff, IDE diffs) or web-based code review platforms (e.g., GitHub, GitLab), seendiff is a persistent, stateful desktop application focused on the process of reviewing, not just the display of differences. It bridges the gap between raw AI output and human comprehension.
2. Key Innovation: The combination of persistent local review state and a first-class integration for AI self-explanation. This creates a novel workflow where the AI agent is not just the author but also an active participant in the review process, directly addressing the context gap between AI-generated code and human reviewer expectations.

Frequently Asked Questions (FAQ)

1. How does seendiff handle merge conflicts or rebased code? seendiff's review state is resilient. It stores progress based on code block content and location, allowing it to survive and re-sync after common Git operations like rebase, amend, or even moving the repository, so you don't lose your review work.
2. Is seendiff a replacement for GitHub or GitLab code review? No, it is a complementary tool. seendiff is for the deep, focused, local review process, especially for huge initial drafts from AI. You would still use GitHub/GitLab for collaboration, comments, and final approval workflows after using seendiff to understand the changes.
3. What are the system requirements for installing seendiff? seendiff requires Python 3.10 or higher and Git to be installed on your system. It is a Python CLI application that can be installed via uv, pip, or from source, and it supports macOS, Linux, and Windows.
4. Does seendiff send my code or review data to the cloud? Absolutely not. seendiff is designed with security and privacy as core principles. It runs a server strictly on localhost (127.0.0.1), does not require an internet connection after installation, and has zero telemetry. All data remains on your machine.
5. Can I use seendiff with any AI coding agent? The integrated walkthrough skill is designed to work with AI agents that have shell access and can execute Python scripts, such as Claude Code. The seendiff CLI itself can be used to review any Git diff, regardless of its origin (AI or human).