AI Agent Task Manager

AI Agent Task Manager is a free browser-based kanban board from Codersera that lets developers run multiple AI coding agents — Claude Code, Codex, and Cursor — in parallel by queueing tasks per workstream and letting agents pull work autonomously.

What is the AI Agent Task Manager?

It is a free browser-based task board for managing several AI coding agents that work on the same or different projects at the same time. Each column of the board is a workstream that maps to one agent session; each card is a task with a title and a full instructions body. The board runs in any modern browser, is made by Codersera, and works with agents like Claude Code, Codex, Cursor, or any tool that can make HTTP requests. Signed-in users get a personal API key so agents can pull tasks from the board and report progress back over a small REST protocol.

Key Features

• Workstreams — Create up to 24 columns per board, each mapped to one autonomous agent session running on your machine.
• Task cards with instructions — Each card has a title plus an instructions body that the agent receives verbatim when it pulls the task; the top card in a queue is the next one worked.
• Copy-paste agent connection — The "Connect an agent" button generates a bootstrap prompt and personal API key; paste it into a fresh Claude Code, Codex, or Cursor session and the agent self-drives: pull → do → complete → loop.
• Attention flags — When an agent needs a human decision, the column turns amber, the browser fires a notification, and with Web Push subscribed your OS gets pinged even when the browser is closed.
• REST protocol — A simple pull / attention / resume / complete API lets any script or agent framework drive the board; long-polling supports idle agents without busy polling.
• Cross-device sync — Anonymous boards are stored in the browser's IndexedDB; signing in with Google persists boards across devices and enables the API key.
• Drag-and-drop reordering — TODO cards can be dragged within or across columns; running and "needs you" cards are locked so an in-flight agent can't be moved sideways.

Who is it for?

• Developers juggling multiple agent sessions — Track five or six Claude Code/Codex/Cursor terminals in one board instead of hunting across terminals for the one that's stuck.
• Solo founders running overnight work — Queue 15 small chores on one workstream, paste the bootstrap prompt before logging off, and get done tasks plus result summaries by morning.
• Teams doing parallel feature work — Split frontend, API, and test work into three workstreams so each agent picks up its own queue without blocking the others.
• Engineers experimenting with self-hosted models — Anything that can run curl (Ollama, vLLM, custom Python scripts) can drive the board, since it only speaks plain HTTP.

What can you do with it?

• Parallel feature work — Run front-end, API, and test agents simultaneously; when one gets blocked on a migration decision, its column lights up amber while the others keep running.
• Overnight backlog grind — Let an agent work through a stack of typo fixes, doc updates, dependency bumps, and low-risk refactors, flagging only the ones that need a real call.
• Big refactor while shipping — Use one workstream for a slow refactor and another for fast feature work on a separate branch, with separate queues per stream.
• Manual board without automation — Use it as a plain kanban board to track what each parallel session is doing, without ever connecting an agent.

How does it work?

Create a workstream per agent session, add tasks with titles and instruction bodies, and click "Connect an agent" to generate an API key and bootstrap prompt. Paste the prompt into a fresh agent terminal; the agent then claims the top task, works it, posts a one-line result on completion, and loops back for the next. When it needs a decision, it raises an attention flag, the column turns amber, and your browser notifies you.

Pricing

The board is free. No signup is required to try it; signing in with Google is free and unlocks cross-device sync plus the personal API key for autonomous agents. Codersera states there are no artificial caps on the free tier, no credit card, and no training on user data.

Alternatives

• Notion / Google Docs — fine as a scratchpad of what each session is doing, but with no API for agents to read, status drifts quickly.
• Trello / Linear / Jira — built for human assignees and manual status changes; their API surface is geared to integrations, not autonomous polling.
• tmux / Zellij — let you see every session in one screen but don't tell you which one needs you.

FAQ

Is it free, and do I need to sign up?

The board is free. You can use it anonymously with no signup — it saves to your browser via IndexedDB. To connect autonomous agents (which need a private API key) and sync your board across devices, sign in with Google.

How do agents connect to the board?

Click "Connect an agent" on any workstream, generate your personal API key, and copy the bootstrap prompt it gives you. Paste that into a fresh Claude Code, Codex, or Cursor session. The prompt tells the agent to call the board over HTTPS: pull its next task, do it, mark it complete, and repeat.

What happens when an agent gets stuck and needs me?

The agent flags the task, that column turns amber ("needs you"), and your browser fires a notification. You answer the agent at that terminal; it then clears the flag and resumes itself. The board is an at-a-glance signal, not a chat relay.

Is my data private?

Anonymous boards never leave your browser. Signed-in boards are stored in Codersera's database scoped to your account. Codersera says it never shares or sells data and never uses it to train models; task instructions are stored as plain text so agents and you can read them back exactly.

Which agents and tools work with it?

Anything that can make an HTTP request. It is built and tested for Claude Code, OpenAI Codex, and Cursor in full-auto mode, but the protocol is plain REST, so your own scripts, self-hosted Claude via the Anthropic API, OpenAI-compatible local models via vLLM, Ollama agents, or custom Python scripts can drive it too.