Plannotator is a free, open-source review user interface specifically designed for AI coding agents. Developed by backnotprop, it was initially announced on December 27, 2025, with the goal of empowering developers to maintain ownership and control over the outputs of their AI coding assistants. This plugin integrates natively with various agents such as Claude Code, OpenCode, Codex, and Pi, offering a robust platform for supervising and refining AI-generated development work. The tool operates entirely locally, ensuring that sensitive plans and code never leave the user's machine, thereby upholding privacy and security standards in AI-assisted development workflows.

Plannotator serves as a critical human-in-the-loop mechanism by intercepting the workflow of AI coding agents at key stages: plan generation and code modification. When an AI agent proposes a plan for a task, Plannotator automatically presents this plan in a rich, browser-based review environment, moving beyond simple terminal text. Similarly, for agent-written code, it provides a comprehensive diff viewer for uncommitted changes or pull requests. Users can then interact with these outputs directly by adding inline annotations, marking sections for deletion, adding comments, or suggesting replacements. This annotated feedback is then converted into a structured format that the AI agent can interpret and use for precise revisions, fostering an iterative refinement loop that ensures the agent's output aligns with the developer's intent.

Some of the key features are:

• Plan Review: Automatically intercepts an AI agent's proposed plan, presenting it in a dedicated UI for detailed human review before execution.
• Code Review: Provides a PR-style diff viewer to inspect agent-generated uncommitted code changes or external GitHub/GitLab pull requests.
• Structured Feedback: Allows users to annotate specific parts of plans or code with comments, deletions, or suggestions, which are then converted into structured feedback for the agent.
• Inline Comments: Add comments on any section of the plan or specific lines of code within the diff viewer.
• Deletion Marking: Mark sections of text in plans or code for removal, providing clear instructions for the agent to reduce scope or refactor.
• Version History: Tracks every revision of a plan, providing visual and raw markdown diffs to highlight changes between iterations.
• Local Execution: Operates entirely on the user's local machine, ensuring that sensitive plans and code remain private and never leave the environment.
• Encrypted Sharing: Enables secure sharing of review sessions via URL, where all data is contained within the link itself, eliminating the need for a backend or user accounts.
• Draft Auto-save: Automatically saves annotations and review drafts to prevent loss of progress due to crashes or restarts.
• VS Code Integration: Facilitates opening plans in VS Code editor tabs, utilizing its diff view and allowing for editor-based annotations.
• Obsidian/Bear Integration: Offers automatic saving of plans to Obsidian vaults or Bear notes, including frontmatter and tags for better organization and searchability.
• Image Attachments: Supports copying and pasting images or uploading files directly into annotations, providing visual context such as mockups, screenshots, or diagrams.
• AI Chat (Code Review): Integrates inline AI chat within the diff viewer, allowing users to ask questions about selected code lines and receive streaming responses.
• Multiple Diff Types: Provides various options for comparing code, including uncommitted, staged, unstaged, last commit, and comparison against a default branch, along with Jujutsu (jj) specific diff modes.
• Web Page Annotation: The /plannotator-annotate command can fetch and convert URLs or local HTML files into markdown for annotation, enabling structured feedback on external documentation.
• Agent Switching (OpenCode): Allows OpenCode users to configure which specific agent receives the approved plan, enhancing flexibility in multi-agent workflows.

Plannotator functions by hooking into the workflow of supported AI coding agents. For plan review, when an agent reaches its ExitPlanMode or Stop hook (depending on the agent), Plannotator intercepts this approval step. It launches a local web server and opens a browser window displaying the agent's proposed plan. Users can then interact with this UI to add annotations. Once the user approves or denies with feedback, the annotations are serialized and sent back to the agent as structured input, prompting it to refine its plan. For code review, users explicitly invoke the /plannotator-review command. This command instructs Plannotator to capture local git diff changes or fetch the diff of a specified GitHub/GitLab pull request. A browser-based diff viewer is then presented, allowing line-level annotations, staging/unstaging files, and inline AI queries. Upon submission, the structured feedback or approval message is passed back to the agent for further action, such as committing changes or making revisions. The tool respects environment variables for configuration and offers CLI options for installation and version pinning.

Some common use cases include:

• Refining AI-Generated Plans: Developers can review and guide an AI agent's development strategy before any code is committed, ensuring the plan aligns with architectural guidelines and project goals.
• Quality Assurance for AI Code: Reviewing AI-written code with a familiar diff viewer and providing precise, line-level feedback to ensure code quality, adherence to standards, and correctness before integration.
• Collaborative AI Development Workflows: Sharing agent-generated plans or code diffs with teammates via unique URLs for collaborative review and feedback, even without a shared backend system.
• Integrating External Documentation into AI Prompts: Using the annotate command with URLs or local HTML files to feed specific, highlighted sections of documentation directly to AI agents, ensuring they work with the correct context.
• Improving Agent Reliability: Analyzing plan revisions and feedback patterns over time to continuously improve the performance and reliability of AI coding agents, learning from past denials and successful approvals.
• Onboarding and Training: Utilizing Plannotator to walk through an agent's planned steps and code, providing real-time feedback that can serve as a learning and training mechanism for both the developer and the AI.
• Secure Local Development: Performing sensitive code and plan reviews entirely locally, preventing proprietary information from being exposed to external services, which is crucial for enterprise environments.