Claude Code, with a real interface

The first thing to be clear about, because it is the thing people assume otherwise: AIDEN does not ship its own agent. It starts the Claude Code CLI that is already on your machine, the one you installed yourself, authenticated yourself, and pay for yourself. AIDEN passes it a working directory and an approved spec, then reads its output into a UI. That is the whole relationship.

Which means the boring reassurances are all true by construction. Your login is your login, so there is no second account to create and no key to paste into a settings pane. Your usage limits are your usage limits, unchanged; AIDEN does not make the model faster, cheaper, or more permissive. Your CLAUDE.md, your MCP servers, and your skills are already configured in the CLI, so they apply exactly as they do in a terminal. Remove AIDEN tomorrow and your CLI is untouched.

This is also the honest answer to "so it is another Cursor?". It is not an editor with a chat panel; it is the layer above the CLI that decides what gets built, where it gets built, and who approves it. The general shape of that layer is described in what an agentic development environment is. AIDEN is one, built specifically around the Claude Code and Codex CLIs.

A terminal is an excellent interface for one agent doing one thing while you watch. It gets worse with every agent you add, and not because the CLI is bad, but because a terminal shows you the most recent output of the one tab you have focused. Four specific things fall out of view.

The worktree point is the one that stops being a nicety and starts being load-bearing fastest. Two Claude Code sessions pointed at the same checkout overwrite each other's files, stage each other's changes, and fight over the same ports; the mechanics and the manual recipe are in running parallel agents with git worktrees. The case for the board specifically, and what a card has to track to be worth anything, is in a kanban board for Claude Code.

The board has a gate in it, and it is the part of AIDEN that is a process decision rather than a UI one. A story does not go straight to an agent. It goes to Spec Review first, where AIDEN drafts a written plan from it, goal, acceptance criteria, boundaries, and the card sits there until you edit and approve it. Only then does an agent get a worktree and start writing code.

1. 1

   Write the story

   One deployable slice, in plain language. "Add rate limiting to /api/checkout" produces a spec an agent can be held to; "improve the API" does not.
2. 2

   Read the drafted spec

   AIDEN turns the story into goal, acceptance criteria, and explicit boundaries. This is the cheapest point in the whole cycle to catch a misunderstanding, before any code exists to argue with.
3. 3

   Approve, and the agent starts

   The card moves to In Progress on its own worktree and branch, with the approved spec as the agent's context. Run one card or five; the gate is the same.
4. 4

   Review the diff against the spec

   Finished cards arrive with the diff next to the contract it was supposed to satisfy, so review is a comparison rather than an act of reverse-engineering intent.

The reason this matters more with a GUI than without one is arithmetic. The moment you can run five agents comfortably, your bottleneck moves from writing code to reviewing it, and an unreviewable diff is worse than no diff. The gate is what keeps the number of agents you can run tied to the number of specs you are willing to approve. The orchestration underneath it, how stories queue and how branches become PRs, is in Claude Code orchestration.

AIDEN is bring-your-own inference. You keep the Claude Code subscription or API key you already have, and AIDEN runs on top of it. It never resells tokens and adds no model cost of its own; what you pay AIDEN is for the workspace, and what you pay Anthropic is for the model, exactly as you do today.

On keys and code, the answers are the same ones on our homepage FAQ, word for word, because they are the questions that decide whether a desktop app is installable at all:

The same applies to Codex if you run both. Cards are assigned to your Claude Code CLI or your Codex CLI, on one board, behind one spec gate, with one review flow, which is the practical reason a vendor-neutral interface exists at all. If you are weighing the field, the other desktop apps in this category, including Anthropic's own, are compared in desktop apps for coding agents.

The platform situation is the one place this page has to be blunt, because downloading a build that does not exist for your OS is a bad first experience.

We are not putting a date on the Windows build. Store certification is not a timeline we control, and a missed date costs more with this audience than a vague one does. What we can say is that it is in certification rather than on a roadmap, and that everyone on the waitlist gets an email the day it lands. Pick your platform there and that is the whole commitment, no newsletter.