AgentDeck
Sign in Live demo
Early access

Hand your project to an AI agent. Watch it work.

AgentDeck is mission control for Claude. Connect Claude, upload a project, and AgentDeck builds the project's memory, turns your instructions into commands and runs the agent, with chat, terminal and task board in one browser tab.

Try the live demo See how it works

No sign-up. The demo runs entirely in your browser.

Works with the code you already have
  • TypeScript
  • JavaScript
  • Python
  • PHP & Laravel
  • React
  • Node.js
  • Go
  • SQL

How it works

From upload to working agent in four steps

  1. 1

    Connect Claude

    Add AgentDeck as a connector in your Claude app and use your own Claude plan, or add an Anthropic API key for the built-in agent. Both take a minute.

  2. 2

    Upload your project

    Drop in a folder or zip, or connect a Git repo. Each project gets its own private workspace.

  3. 3

    Memory builds itself

    AgentDeck reads the code and writes a CLAUDE.md: the stack, how to run and test it, where things live, and your rules.

  4. 4

    Give instructions

    Say what you want in plain words, or use commands like /test and /review. The agent plans, edits, runs tests and reports back.

Features

Everything your agent does, in one place

Live chat with the agent

See every step as it happens: which files it reads, what it changes, and what it runs.

Real terminal

The same shell the agent uses. Check its work or run your own commands at any time.

Project memory

A CLAUDE.md plus saved facts you can edit. The agent remembers your project between sessions.

Task board

Every plan turns into tasks you can follow from To do to Done.

Context meter

See how much of the agent's working memory is used, and pick up the same session after a reconnect.

Your own commands

Save instructions you repeat as one-word commands, like /test, /cleanup or /release-notes.

Two ways to connect

Works with the Claude you already pay for

Your Claude app + AgentDeck connector

Add AgentDeck in Claude (Pro, Max or Team) or Claude Code, sign in once and press Allow. Claude then works on your projects here, on your own plan: it lists projects, reads and edits files, keeps the memory and runs your tests. Your projects stay on the always-on workspace.

Settings → Connectors → Add custom connector → paste your AgentDeck URL.

Built-in agent with your API key

Add an Anthropic API key and AgentDeck's own agent works inside this dashboard: live steps, diffs, approvals, usage per run, and tasks that keep running after you close the tab.

Keys are stored encrypted and used only for your projects.

Safety

Nothing gets deleted without your OK

An agent that can run commands needs guardrails. AgentDeck stops and asks before anything you can't undo.

  • Approvals for risky actions. Deleting files or data, database changes and deploys wait for your click.
  • A private workspace per project. The agent works in its own container, with no access to your other projects or servers.
  • Your key stays yours. Stored encrypted, never shared, and you can remove it at any time.
  • A full activity log. Every command and file change is recorded, so you can see exactly what happened.
Approval needed

The agent wants to delete data.

DELETE FROM sessions
WHERE created_at < NOW() - INTERVAL 30 DAY;
Rows
1,284
Database
shop (production)
Backup
Today, 03:00
Deny Approve

FAQ

Questions

How does AgentDeck connect to Claude?

Two ways. Add AgentDeck as a connector in your Claude app or Claude Code and work on your projects with your own Claude plan. Or add an Anthropic API key (console.anthropic.com) and AgentDeck's built-in agent runs in the dashboard, billed by Anthropic to your account. You can use both.

What kind of projects work?

Any code project with files and a terminal: web apps, APIs, scripts, Laravel or WordPress sites, data jobs. Big generated folders like node_modules are skipped automatically.

Can the agent break things?

It works on its own copy of your project, and risky actions need your approval first. You review the changes, then download them or push them to Git when you're happy.

Is my code private?

Each project lives in its own workspace. Code is sent to Claude through Anthropic's API only when the agent needs it for your task.

Is the demo real?

The demo is a simulation that runs entirely in your browser. If you pick your own folder, it's read on your device to build the memory and run the read-only checks. Nothing is uploaded. The agent's code changes in the demo are scripted.

Is AgentDeck made by Anthropic?

No. AgentDeck is an independent product that uses Claude through Anthropic's API.

See it on your own project

Open the demo, pick a folder, and watch AgentDeck build its memory, right in your browser.

Open the live demo