First Launch

Everything that happens after you install the app: macOS permissions, AI configuration, and you are ready to run your first session.

Before reading this, make sure you have completed Prerequisites (calendar setup, and cloud credentials only if you want a cloud engine).

1. Grant macOS permissions#

On first launch, Azynote will ask for four macOS permissions. Grant all of them for the full experience.

Microphone#

Azynote transcribes your voice locally via Whisper during sessions. When macOS prompts you, click OK / Allow.

If you missed the prompt:

1. System Settings > Privacy & Security > Microphone
2. Toggle Azynote on
3. Restart Azynote

Calendar#

Azynote displays your meetings and links sessions to calendar events. When macOS prompts you, click OK / Allow.

If you missed the prompt:

1. System Settings > Privacy & Security > Calendar
2. Toggle Azynote on
3. Restart Azynote

You can verify the status inside the app: Settings > Calendar. The macOS Calendar row should read "Access granted". If it doesn't, click Refresh.

Screen & System Audio Recording#

Azynote captures meeting app audio separately from your microphone, so it can tell who said what. macOS will show a Screen & System Audio Recording prompt. Click Allow.

If you missed the prompt:

1. System Settings > Privacy & Security > Screen & System Audio Recording
2. Toggle Azynote on
3. Restart Azynote

Accessibility#

Azynote registers global hotkeys that work from any app: Quick Capture (Control+Shift+K), the Command Panel (Control+Shift+O), dictation (Control+Shift+D) and voice conversation (Control+Shift+V). Dictation also types into the app you are in. All of this requires Accessibility access. macOS will prompt you. Click Open System Settings, then toggle Azynote on.

If you missed the prompt:

1. System Settings > Privacy & Security > Accessibility
2. Toggle Azynote on
3. Restart Azynote

2. Choose your AI engine#

When you open your first workspace, Azynote shows a Choose your AI engine screen with five cards. You have to pick one, there is no skip: an unspoken choice is how people end up wondering why nothing works.

On a Mac with 24 GB of memory or more, Azynote pre-selects Local. On a 16 GB Mac it pre-selects Gemini instead, and the Local card tells you why: on that Mac Local suits Chat and the Command Panel, while OnePagers run better on a cloud engine. You can still pick Local. You can change engines later in Settings > AI engines.

Local (the default)#

Nothing to configure. Click Local, confirm, and Azynote downloads a model sized for your Mac, with a progress bar. When the download finishes, you land in your workspace and everything runs on your Mac.

Needs 16 GB of memory or more. Azynote shows a warning on the card if your Mac has less.

Local models#

Local is a library, not one model. Open Settings > AI engines and click Local in the providers at the top to see it. Models are grouped under Fits this Mac and Needs more memory than this Mac has. Each card says what the model is good at and what it is not, about how much memory it holds on your Mac, how fast it is compared with the others (Fastest, Fast, Medium speed or Slower than the others), and which rows use it. Download or delete a model from its card. A model that a Local row uses cannot be deleted: pick another model for that row first.

Once a model is downloaded, it shows up in the engine menu of each row as Local · model name. That lets OnePager run on one model while Chat or the Command Panel runs on another. A row shows a short note when its model is weak at that job.

Several local models can be loaded at once. How much memory that takes is your call, and an In use line on the Local page shows about how much memory your local choices use out of your Mac's total. Keep local models ready loads the models your Local rows use when Azynote starts. With it off, each model loads the first time it is used and is released after a few minutes idle.

Gemini#

1. Click the Gemini card
2. Paste your API key from Prerequisites into the key field
3. Confirm

If you ever see "Invalid API key", regenerate the key in AI Studio and paste the new one in. To use Vertex AI instead of an API key, pick Gemini here, then switch to Vertex in Settings > AI engines and fill in your GCP Project ID and Region. The ADC Status indicator should show a green checkmark.

OpenAI#

Two choices on the card:

• OpenAI: paste your API key from platform.openai.com and confirm. That is all: Azynote writes with GPT-6 Sol and talks live with GPT Realtime mini, and you can change either later in Settings > AI engines, Providers, OpenAI.
• Compatible server: for OpenRouter, Mistral, Groq, or a server on your Mac like Ollama or LM Studio. Fill in the base URL (for example http://localhost:11434/v1), the model it serves, and an API key if it needs one. Compatible servers handle text; voice then runs turn by turn, with Whisper on your Mac.

CLI#

If you already run an agent in your terminal, Azynote spawns your command and routes AI work through it. The default command targets Claude Code, and you can point it at anything else you run. The card tells you whether it found the command, and a Test command button runs it once and shows you exactly what was spawned and what came back.

In-app chat is switched off with this engine: you chat from your own terminal through the workspace's MCP integration instead. Use ACP below if you want the chat panel.

ACP#

ACP reaches the same kind of agent the CLI engine does, with one difference that matters: the agent is started once and stays running, so every request after the first skips its startup entirely. The slower your agent is to wake up, the more that saves.

It also keeps the in-app chat panel, powered by your own agent, which the CLI engine cannot do.

You give it a command that launches your agent, with no placeholders, because the protocol negotiates the prompt, the model and Azynote's tools itself. Examples: goose acp, gemini --acp, qwen --acp, cortex acp serve.

Not every agent speaks ACP. Azynote never asks your agent for file or shell access over this connection, and your agent's sign-in stays its own business.

Using CoCo CLI or Claude Code#

Step-by-step pages cover the two agents people ask about most: Use CoCo CLI and Use Claude Code. Claude Code needs a small adapter first, and that page walks you through it.

Next: Core Concepts. Read it once. The rest of the app will click.