Local agentic development environment

"Local" gets used loosely, so it is worth being precise. In a local agentic development environment, four things live on your machine: the agent runtime that executes the work, your credentials for the model provider, your git repository, and the working files the agent edits. The orchestration layer, the board, the spec gate, the branch management, the review pass, runs on your machine too and drives those agents.

The contrast is a cloud agent, where you hand a tool access to your repository and it clones your code into its own infrastructure to run tasks there. Both can produce good diffs. The difference is custody: in the local model, no third-party orchestration vendor ever holds a copy of your source. To place this in the wider picture, an ADE is the whole workspace around agents; see what is an agentic development environment. Local is a property that ADE can have.

Running agents locally is not just an ideological preference. It changes four concrete things about cost, speed, security, and control.

BYOK means the tool drives your model account rather than reselling access bundled into its subscription. It matters more than it sounds, because inference is the expensive part of agentic coding, and who pays for it, at what price, defines the whole cost model.

With a BYOK ADE like AIDEN, you install Claude Code or Codex, sign in with your own Anthropic or OpenAI account, and the environment runs tasks through those credentials. You pay the provider directly, at cost. The ADE charges for the workflow, the board, the worktree isolation, the specs, the review, not for tokens. That separation keeps the tool honest: it has no incentive to make your agents chatty.

Cloud agents are not a mistake. They trade custody for convenience, and for some teams that is the right trade. Here is the comparison without the spin.

If a hosted engineer that lives entirely in the cloud is what you are weighing, the head-to-head is in AIDEN vs Devin. For the full field with honest trade-offs, see the best agentic IDEs in 2026.

The setup is short because the heavy pieces, the models, are things you may already have installed. AIDEN is the orchestration layer on top.

1. 1

   Install at least one agent CLI

   AIDEN runs on your local Claude Code (Anthropic) or Codex CLI (OpenAI). Install either or both. Your keys stay in ~/.claude and ~/.codex; AIDEN inherits them without reading, copying, or transmitting them.

   npm install -g @anthropic-ai/claude-code
   # and/or
   npm install -g @openai/codex

2. 2

   Download AIDEN for macOS

   Grab the .dmg from aidenapp.org and drag it to Applications. AIDEN runs on macOS 12+ (Apple Silicon and Intel, signed and notarized). Nothing about your code or keys is uploaded during install or use.
3. 3

   Open a local repository

   Point AIDEN at a git repo on your disk. It analyzes the codebase into a technical and business overview locally, detects your installed CLIs, and inherits your Claude Code MCP configuration automatically.
4. 4

   Ship a story on its own worktree

   Create a story, approve the drafted spec, and the agent works on an isolated git worktree on your machine. The mechanics are in parallel agents with git worktrees, and the PR is one click from the finished card.

The desktop app covers an individual developer. When a team needs the orchestration layer on shared infrastructure behind its own network, AIDEN ships a self-host server mode with a Linux installer and a Docker image. The board, the specs, and the review layer run on hardware you control, the agents still execute against your keys, and the code stays inside your environment.

This is the configuration teams reach for when policy forbids source code touching any external SaaS at all. It keeps the parallel-agent workflow, the specs, the worktrees, the PRs, while putting the whole orchestration layer inside your perimeter. The remaining outbound traffic is the model calls themselves, governed by your provider agreement, the same calls your developers already make.