Connect
Connect a model to Growth OX
Three ways to give Growth OX an AI: a local model sized for a 16 GB or 32 GB Mac, an API key kept in the Keychain, or a command-line tool you already signed in to.
The three routes#
Growth OX does not sell AI usage. Each provider's own terms and prices apply to you, and Growth OX never signs in to a provider for you or stores a provider token. The three routes are equal choices, and you can use more than one and give each job to the model you prefer.
| Route | Cost | Where your text goes | Good for |
|---|---|---|---|
| Local model | Free; uses your Mac | Nowhere. It runs on this computer and sends no text over the internet. | Private work, no internet, no per-use bill. Slower and weaker than the largest cloud models. |
| Your own API key | Your provider's price | Straight from your Mac to that provider, only when you ask for a draft. | Strong models, pay for what you use. |
| A command-line tool you signed in to (Claude Code, Codex CLI) | Your existing subscription | To that provider, through the tool you installed. | Using a subscription you already pay for. |
Route 1: a local model#
Local models run through Ollama, which must be installed and running on the same Mac. If Ollama is set to another computer, Growth OX will not install a local model, because a local model only runs on this computer.
- Install and open Ollama.
- In Growth OX choose Measure my computer. It takes a few seconds and downloads nothing.
- Read the verdict next to each model: fits, borderline or does not fit. Models that fit are marked recommended.
- Accept the model's licence, then choose Install. The download shows its progress and you can stop it.
- Try the model with the three short jobs the app offers (a Turkish summary, an English translation and a tool call), then pick which jobs it takes.
A model that cannot call tools is installed but gets no assistant jobs. The app tells you, per model, whether it can.
| Mac memory | Model to start with | Download size | Licence |
|---|---|---|---|
| 16 GB | Qwen3 8B (recommended); Qwen3 4B is a lighter choice | 5.2 GB; 2.5 GB | Apache-2.0 |
| 24 to 32 GB | gpt-oss 20B, or Qwen3 30B (A3B) | 14 GB; 19 GB | Apache-2.0 |
| 48 GB and more | Qwen3 32B or Gemma 4 31B | 20 GB; 19 GB | Apache-2.0 |
On a 16 GB Mac the app marks 20B-class models as too large. Models whose memory need cannot be known before the download show "not known yet" until the app has read them. Speed is not promised: the app shows a measured speed after you run the trial.
Route 2: your own API key#
Choose Add a credential, pick the provider, and paste the key. The key is stored only in this Mac's Keychain; it does not reach the app's window or the browser. The provider list includes Anthropic (Claude), OpenAI, Google Gemini, Mistral, DeepSeek, OpenRouter and others, plus Ollama and LM Studio on this computer.
- Use Test to check that the key works. The messages say whether the key is invalid, lacks permission, has no credit, or the provider cannot be reached.
- Each credential can have a monthly limit. At 80 percent of it you get a warning, and at the limit the job stops.
- Each credential carries a data class (public data only, up to internal data, or up to personal data), so a key you trust less is not used for sensitive jobs.
- When you delete a credential, also revoke the key in the provider's console. Deleting it in Growth OX does not do that for you.
Route 3: a tool you already signed in to#
Growth OX can run the Claude Code or Codex CLI you installed and signed in to yourself. Sign-in happens in Terminal and the app never sees your password. Claude gives the best quality, and web research and the Chrome hand-over work with it. If the CLI is signed in with a paid API key instead of the subscription, the app does not use that sign-in.
Gmail is read through the Claude connector, so Gmail needs Claude. Gmail is not supported for Codex yet.
Questions#
- Does Growth OX include AI?
- No. You connect your own: an API key, a local model, or an official tool such as Claude Code or Codex CLI that you installed and signed in to yourself. Each provider's own terms and prices apply.
- Where is my API key stored?
- Only in your Mac's Keychain. It does not reach the app's window or the browser.
- Which local model should I start with on a 16 GB Mac?
- The app recommends Qwen3 8B on a 16 GB Mac and marks 20B-class models as too large. Run the measurement first: the verdict on your own Mac decides.
- Can I use more than one model?
- Yes. You choose which model does which job, for example chat, drafts, summaries, translation or research.
- Does a local model need internet?
- Only to download the model and for jobs that need the web. Once installed, the model runs on your Mac and sends no text over the internet.