Agent Cockpit

Agent Cockpit provides a single control room for managing multiple AI coding agents such as Claude Code and Codex. It aggregates approval requests, diffs, tool calls, and execution state into a unified interface, allowing developers to approve or deny risky actions, inspect changes, and monitor progress without switching terminals. The system records a chronological timeline of events, making it possible to replay sessions and understand the sequence of edits, approvals, and tool invocations.

The platform is built with a three‑layer architecture: a provider layer that emits standardized events from each coding agent, a local Node daemon that normalizes those events, stores them in SQLite, and broadcasts updates via WebSocket, and a React‑based browser UI that displays the live control room. This design keeps all data local, preserving project confidentiality while offering auditability.

Agent Cockpit is open source under the MIT license, written in TypeScript and React, and can be installed and operational within a minute. It targets developers who run several AI‑driven coding assistants simultaneously and need a shared timeline, centralized approvals, and reduced context‑switching overhead.