Handler

Key Features

• •Built-in explanations: Every AI-proposed edit arrives with a natural-language summary of why it was made, what files it touches, and how it fits the task, enabling developers to understand changes before accepting them.
• •Side chat for individual edits: Each diff has a dedicated chat thread where developers can ask questions or request changes without polluting the main agent’s context, keeping the review focused and asynchronous.
• •Agent reads the terminal: The agent can capture terminal output, expand JSON structures, and highlight the exact error line, eliminating the need to manually paste logs and speeding up debugging.
• •Model switching without context loss: Swap between Codex and OpenCode mid-session while preserving all conversation history, allowing developers to leverage different models’ strengths seamlessly.
• •Screenshot integration: Screenshot any part of the screen and hand it directly to the agent via the command bar, useful for referencing UI bugs or design mockups during code generation.
• •Isolated worktrees: Spin up separate Git worktrees for each agent so they never interfere with each other’s files, enabling parallel experimentation without risking the main branch.
• •Multiple agent tabs: Run several agents at once in a clean tabbed interface, each with its own context and history, allowing efficient multitasking across different tasks or projects.

Use Cases

• •Reviewing a large AI-generated refactoring: When an agent proposes a 600-line diff, Handler breaks it into individual edits with explanations. The developer questions each change in side chats, ensuring full understanding before merging into the main branch.
• •Debugging a failed test: After an agent writes code that causes a test failure, the terminal reader shows the exact error line and relevant log data. The agent analyzes the output and suggests a precise fix, reducing back-and-forth.
• •Switching models mid-task: A developer starts a feature with Codex but wants to try OpenCode for the final styling. They switch models without losing the conversation context, preserving all previous instructions and history.
• •Incorporating design feedback: A designer screenshots a UI mockup and hands it to the agent via the command bar. The agent generates the corresponding frontend code, matching the visual design exactly.
• •Running multiple agent experiments: An engineering team spins up isolated worktrees for each agent to test different refactoring approaches. Agents run in parallel without file conflicts, and only approved changes are merged after review.
• •Onboarding new team members: New developers use Handler to review AI-generated code with explanations, quickly understanding the agent’s logic and learning the codebase’s conventions without needing a senior engineer to explain every change.

Who is this for?

Handler is designed for individual developers and engineering teams who rely on AI coding agents such as GitHub Copilot, Codex, or OpenCode and want to maintain code quality and understanding. It is ideal for senior engineers reviewing AI-generated pull requests, tech leads ensuring accountability in AI-assisted development, and teams experimenting with multiple agent models. The tool also suits freelance developers who need a rigorous review process to confidently ship code to clients. Handler targets those who value transparency, seek to avoid blind acceptance of AI output, and want a structured workflow that combines the speed of AI with the discipline of traditional code review.