Five-minute setup

Choose how CueDraft thinks.

CueDraft discovers supported providers already available on your Mac. Pick one that is ready, review your shortcuts, and start using selected-text actions.

Open CueDraft

Move CueDraft to Applications and open it. The onboarding flow is resumable, so you can close it and continue later.

Allow Accessibility

Open System Settings → Privacy & Security → Accessibility and enable CueDraft. The status refreshes immediately. You may defer this step, but automatic selection capture, Quick Fix, and replacement need the permission.

Install or connect one provider

Ollama

Install Ollama, download one or more models, and keep the default local server or configure another URL. Choose the model you want in CueDraft; a remote URL sends selected text to that server.

Official Ollama setup ↗

Claude CLI

Install Claude Code and complete its own sign-in. CueDraft reuses that CLI login and does not ask for the provider credential.

Official Claude Code setup ↗

Codex CLI

Install Codex CLI and sign in with its supported account flow. CueDraft reuses that CLI login without storing it.

Official Codex CLI sign-in ↗

Cursor Agent

Install Cursor Agent and sign in through its own account flow. CueDraft discovers the command for the current Mac user and never asks for a Cursor API key.

Official Cursor Agent setup ↗

Gemini CLI

Install Gemini CLI, run it once, and choose Sign in with Google. CueDraft uses that cached sign-in and deliberately ignores Gemini and Google API-key environment variables.

Official Gemini CLI setup ↗

Test readiness

Choose a provider in onboarding. CueDraft checks that Ollama has a model or that the selected CLI is available. At least one provider must be ready before setup can finish.

Review shortcuts

Quick Command: Shift–Command–Space
Quick Fix: Control–Shift–J
Open CueDraft: Control–Command–C

Every shortcut is configurable in General Settings, and CueDraft prevents conflicts between its actions.

Fast, isolated requests

Every action starts a fresh provider request. Claude, Codex, Cursor, and Gemini do not reuse a prior CueDraft conversation. CueDraft minimizes reasoning where each provider supports it. Ollama’s permanent keep_alive keeps your selected model loaded for speed; it does not carry prompt history into the next request.