Agentic is in beta — some features may not work as expected.

Agentic docs

Everything you need to know about using Agentic: how to install it, which AI providers it supports, the permissions it needs, and answers to common questions.

How to use Agentic

  1. Install

    Download Agentic.dmg, double-click it to open the disk image, and drag Agentic into the Applications folder that appears inside it. Then open Agentic from your Applications folder.

    Because the app is signed for local use, macOS may ask you to open it the first time: right-click (or Control-click) the app and choose Open, then confirm.

  2. Grant permissions once

    Open Settings → Permissions in the app and press each Request button. This is the only step you do by hand — after that the app never asks again:

    • Screen Recording — so it can see your screen (only used when you ask it to look).
    • Accessibility — so it can click, type, and press keys for you.
    • Notifications — so it can show a banner and play its custom sound when your answer is ready.
    • Files and Folders — macOS grants this automatically the first time it touches Desktop, Documents, or Downloads.

  3. Choose an AI provider

    Agentic starts on the Local (Ollama) model, which is private, free, and downloads itself on first use. In Settings → General you can switch to Gemini, OpenRouter, Groq, or OpenAI and paste a free API key from the shown link (OpenAI is billed per token).

  4. Just talk to it

    Type normally, the way you would ask a person:

    • "What is the latest version of macOS?"
    • "Search my Downloads folder for the newest PDF."
    • "Open YouTube in Chrome and play a video about how computers work."
    • "Remind me in 15 minutes to take a break."
    • "Write a to-do list for today to a file called todo.txt."
    • "List the apps that are open right now."

    While it works, you see each step it is taking in the activity log, above its live response.

  5. Approve the big actions

    Safe actions run automatically. Anything that could change your computer — running a command, writing or editing a file — shows you what it is about to do and waits for your yes or no. Destructive actions (deleting, formatting, sudo) always warn you clearly.

  6. Find your conversations

    Past chats appear in the sidebar on the left and in the menu-bar Conversations menu. Select one to reopen it. Use ⌘N or the toolbar button to start a new conversation.

AI providers

Agentic works with five AI providers. The local model is the default because it is private and free; the cloud providers need a free API key pasted in Settings → AI Provider.

Ollama · local

Runs a model on your own Mac. Private, unlimited, free — and it downloads itself on first use. No key needed. The default.

Google Gemini

Google's Gemini models. Get a free API key at aistudio.google.com/app/apikey.

OpenRouter

One key, many models — including permanently free tiers. Get a key at openrouter.ai/keys.

Groq

Very fast free models. Get a free key at console.groq.com/keys.

OpenAI

ChatGPT models (like gpt-5-mini), billed per token. Uses an API key, not a ChatGPT login — create one at platform.openai.com/api-keys.

About the permissions

Agentic only uses your permissions when you ask it to. It never looks at your screen, clicks, or types on its own — it only acts inside a conversation you started. Every permission can be granted or revoked anytime in System Settings → Privacy & Security, and the app's own Permissions tab always shows the real status with one-tap links to the right settings pane.

Needed for

Screen Recording

Needed only for capture_screen — when you ask it to look at what is on screen.

Needed for

Accessibility

Needed for clicking, typing, and key presses. This is what lets it operate apps for you.

Needed for

Notifications

Lets it play the custom chime and show a banner when your response or reminder is ready. Even if this is off, the in-app sound still plays.

Troubleshooting

A provider keeps erroring

Switch to another provider or the local model in Settings → AI Provider. Free-tier keys can hit daily limits; the local (Ollama) model has no limits.

macOS won't open the app

Right-click (or Control-click) Agentic in Applications, choose Open, then confirm. This is the standard Gatekeeper step for apps that aren't notarized yet — you only do it once.

It can't do something it should

Open Settings → Permissions — it shows the real status of each permission with a button to fix it. Grant anything marked as missing.

Permissions disappeared after an update

They normally survive updates because the app is signed with a stable identity. If macOS ever forgets them, the Permissions tab tells you exactly which ones and how to restore them.

Frequently asked questions

Is it really free?

Yes. The default provider runs a model on your own Mac (Ollama), which costs nothing. The cloud providers offer free API tiers; paste a free key in Settings if you prefer one of them. OpenAI is also supported and is billed per token.

Which AI providers are supported?

Five: Ollama (local, default), Google Gemini, OpenRouter, Groq, and OpenAI. See the AI providers section above for how to get each key.

Is OpenAI supported?

Yes — Agentic now supports OpenAI (ChatGPT models like gpt-5-mini). It uses an OpenAI API key, billed per token on your OpenAI account. Add a key in Settings → AI Provider.

Can I sign in with my ChatGPT account?

Not yet. Agentic connects to OpenAI with an API key rather than your ChatGPT login. An API key is created in a minute at platform.openai.com/api-keys and only sends your messages to OpenAI for processing.

What Macs does it run on?

Agentic needs macOS 14 or later and an Apple Silicon Mac (M1 or newer) — Intel Macs are not supported. It was built and tested on macOS 26.

Why does macOS show a security warning?

Agentic is not notarized by Apple yet, so macOS can't verify its developer when you first open it. It's safe: the app is open source and signed, and your messages stay on your Mac unless you use a cloud provider. To open it, right-click (or Control-click) the app and choose Open, then confirm. You only do this once.

Does it work with VoiceOver?

Yes — VoiceOver support is a core feature. Every step, approval, error, and finished response is announced, buttons and lists have clear labels, and the app plays its own sounds so you know when something happens.

What are subagents?

For bigger jobs, Agentic can delegate one focused piece of work at a time to a subagent — a researcher (investigates and reports, never modifies anything), an implementer (builds or changes things, with your approval), a verifier (checks that work is correct), or a general worker. You can ask for it directly, e.g. "Delegate this to a researcher subagent."

Can it run more than one agent at the same time?

No — Agentic runs a single main assistant. It can delegate to a subagent (researcher, implementer, verifier, or general), but only one at a time: it launches one, waits for its report, then continues. Subagents can't launch other subagents.

Is my data private?

With the local model, your conversation never leaves your Mac. With a cloud provider, your messages are sent to that provider under their privacy policy. Conversations are saved on your Mac in Application Support and can be deleted anytime. See the privacy page for the full picture.

Why does it need a one-time permissions re-grant after updates?

It shouldn't, and normally it won't: Agentic is signed with a stable identity so permissions survive updates. If macOS ever forgets them, the Permissions tab tells you exactly which ones and how to restore them.

Can it do everything for me?

It can do a lot: browse, search, read and write files, run commands, set reminders, and operate on-screen apps. Anything irreversible still stops to ask you first.

Can I ask questions about the app on this website?

Yes — click the Ask Agentic AI button in the corner of any page. It's a small AI assistant that answers questions about Agentic, like how to install it, which AI providers it supports, or how permissions work.

Is Agentic finished?

Not yet — it's in beta. The core features work, but some things may not behave as expected. If you hit a bug, it's being actively fixed.