Connect an AI provider

Add a provider credential so ADE can start agents and worker runs.

Open AI provider settings

Press Cmd+, (macOS) or Ctrl+, (Windows/Linux), or open Settings and select AI Providers.

---

Add one provider first

ADE works best once you have at least one verified provider configured. There are two types:

API key providers

Paste an API key and ADE verifies the connection immediately.

Anthropic (Claude)

Best default — works with chat, worker runs, CTO, and automations. Keys start with sk-ant-.

OpenAI (GPT)

Good for structured output and GPT-family models. Keys start with sk-.

OpenRouter

Route to hundreds of models through one key. Great for cost optimization.

These are the most common providers. Additional API-key providers (e.g. Google Gemini, Mistral, Ollama) are available in the Settings → AI Providers panel.

1. Paste the API key

   Enter the credential into the correct provider field.

2. Save

   Click Save — ADE stores the credential locally in .ade/local.secret.yaml (never committed to git).

3. Verify

   ADE runs a lightweight check to confirm the credential works. A green indicator means you’re ready.

CLI and SDK providers

ADE detects installed CLIs and SDK-backed provider keys automatically.

Claude Code

Install the claude CLI and sign in. ADE detects it automatically.

Codex

Install the codex CLI globally (npm i -g @openai/codex) and authenticate.

Cursor

Add a Cursor API key. ADE uses Cursor’s agent through the Cursor SDK.

Local CLIs and SDK-backed provider keys are checked during onboarding and on each app launch. If you install a CLI or add a provider key after starting ADE, open Settings and select AI Providers, then click Refresh to re-detect.

---

Which provider should I start with?

If you want…	Use this	Why
The most complete ADE experience	Anthropic (Claude)	Works everywhere — chat, worker runs, CTO, automations
You already use Cursor daily	Cursor	ADE integrates through the Cursor SDK with a Cursor API key
Cost flexibility	OpenRouter	Route different agent roles to different models
Privacy / offline work	Ollama (local)	No API key, runs entirely on your machine

ADE is local-first. Provider calls go directly from your machine to the model provider. ADE never proxies your requests through its own servers.