Plannotator, a local, browser-based review surface for AI coding agents, is climbing GitHub's trending list — and it is betting that the bottleneck in agentic coding is no longer generation but review.
The tool plugs into an agent through plugins, hooks, and slash commands. When Claude Code, Codex, Copilot CLI, Gemini CLI, OpenCode, Kiro, Droid, Amp, DeepSeek Harness, or Pi proposes a plan, a browser window opens so you can annotate it and approve or deny; the agent's next turn carries your structured feedback. A separate code-review mode renders local diffs and GitHub or GitLab pull requests with a full file tree and line comments, and lets you stage or unstage files. Feedback goes straight back to the agent.
The newest piece, Plannotator Inbox, extends that to multi-agent work: one local window where several agents leave messages, questions, files, and guided reviews while they keep working. Answering a question does not hold a session open — the reply simply arrives as the asking agent's next turn. Claude Code, Pi, and OpenCode connect through plugins; other agents reach the Inbox over MCP.
The project says it collects no telemetry and keeps plans, diffs, and annotations local by default, and it ships release binaries with SHA256 sidecars and SLSA provenance attestations. It is dual-licensed under Apache-2.0 or MIT and has drawn roughly 9.9k stars, with about 412 added today.




