AgentTempo

AgentTempo is a local, human-first dashboard for coordinating multiple Codex sessions without turning the dashboard itself into another AI agent.

It groups sessions into workstreams, shows what is running or ready for review, and lets you arrange the same session beneath multiple timeline tasks. Everything stays on your Mac.

What it does

• Monitors local Codex session state: Running, Needs you, Result ready, or Unknown.
• Groups sessions into human-defined workstreams with drag and drop.
• Organizes each workstream as an optional vertical task timeline.
• Shows the current prompt for a session without making an AI request.
• Prioritizes active sessions when refreshing.
• Adds unread-result badges to the relevant project and session.
• Sends native macOS banners and a direct alert sound when configured.
• Keeps workstreams, links, ordering, hidden items, and timeline tasks in browser storage.
• Can automatically link newly created Codex sessions from a configurable New project folder.

Privacy and usage

AgentTempo is designed to run locally:

• The web interface runs on localhost:3000.
• The read-only Codex adapter listens on 127.0.0.1:4319.
• Session refreshes read local Codex state only. They do not call an AI model or consume Codex usage.
• AgentTempo does not upload your session data to a hosted service.
• Native notifications are created by the local adapter on macOS.

Requirements

• macOS
• Node.js 22.13 or newer
• Codex Desktop or ChatGPT Desktop with the bundled Codex runtime

Install with your AI agent

Copy this prompt into Codex:

Clone https://github.com/ChrisZhangWG/AgentTempo to a local folder. Check that Node.js 22.13 or newer is available, install the project dependencies, run the test suite, and start AgentTempo with its launcher. Confirm that both http://localhost:3000 and the local adapter are available. Do not upload or expose any local Codex session data.

After installation, the reusable launcher is available through npm run launch. It starts the local services, waits until they are ready, and opens AgentTempo in the browser. The services stop shortly after the last AgentTempo tab is closed.

Optional configuration

By default, the special New project workflow points to:

~/Documents/codex projects/New project

To use another folder, ask your AI agent to start AgentTempo with the AGENTTEMPO_NEW_PROJECT_ROOT environment variable set to the absolute path you want. New Codex sessions observed under that folder can then be linked automatically to the AgentTempo workstream named New project.

Useful project scripts

Script	Purpose
npm run launch	Start the local services and open AgentTempo
npm run dev:live	Run the web interface and local adapter together
npm run sessions:inspect	Print a read-only local session snapshot
npm test	Build and run the rendered-interface tests
npm run lint	Run the code quality checks

Status detection notes

AgentTempo infers state from the local Codex app server and rollout records. A completed Codex turn means a result is ready for review; it does not mean the entire human workstream is complete. Timeline completion remains a manual decision.

License

MIT