AI providers
AIDEN runs on a unified AI runtime that supports several providers side by side. You choose what powers your agents — and you can mix paths per task.
Supported paths
| Path | What you need | Best for |
|---|---|---|
| Claude Code (Claude CLI) | Claude Code installed, your existing subscription | General engineering work, planning, review |
| Codex CLI | Codex CLI installed, your existing subscription | Coding-heavy execution |
| Anthropic API | An Anthropic API key | Running Claude models without the CLI |
| OpenAI API | An OpenAI API key | Running GPT models; also required for voice |
If you already pay for Claude Code or Codex, the CLI paths are usually the best choice: AIDEN drives the CLIs you have installed, so there is no separate API billing.
Configuring providers
- AIDEN auto-detects installed CLIs on launch.
- API keys are added in Settings and stored locally on your machine.
- You can set a default provider and model, and override it per chat or per agent — model selection is available directly in the chat interface.
- The voice orchestrator has its own delegation setting (Auto, Codex, Claude, or OpenAI API) under Settings → Voice.
Choosing models intentionally
Different tasks deserve different tradeoffs between speed, quality, cost, and reasoning depth:
- fast model — lightweight exploration, quick questions, low-risk repeated work
- stronger model — planning, spec writing, review, anything you will build on top of
- cheaper model — bulk or mechanical work where mistakes are cheap to catch
A good setup is boring: pick a reliable default, and only override when the task clearly calls for it.
The real goal
Do not optimize for model-switching theater. Optimize for a workflow that stays effective and predictable — the structure of stories and specs matters more than squeezing the last percent out of model choice.