Connect an AI model

Before connecting

AI is optional. You can use the complete local editor without enabling it. A provider runs the service; a model is the specific system that answers; a connection saves your provider and model choice.

Remote providers need your own API account, an API key and any required billing or credits. An API key is a credential: enter it only in the Mac app’s connection settings, not in a Markdown file or this website.

Choice	What you need	Where requests run	Actions
OpenAI	API key, account access to a suitable model, API billing	OpenAI	Editing, structure, metadata/tasks, review and Ask
OpenRouter	API key, balance and an eligible catalog model	OpenRouter and its selected provider	Actions supported by the tested model
Ollama	Ollama running locally and an installed model	This Mac’s local Ollama service	Grounded Ask

There is no direct Anthropic, Gemini or Kimi connection in this setup; they appear greyed out in the Provider list. A model offered through OpenRouter must appear in the app’s compatible catalog and pass its connection test.

Enable the optional feature

1. Open Preferences… with ⌘, and find AI Connections.
2. Turn on Enable AI Revision & Review.
3. Read the AI use notice. If you meet the conditions, tick the 18+ acknowledgement and choose Continue to AI connections. This applies to local Ollama as well as remote providers.
4. Open Add your first AI connection and follow the path for your provider.

You do not need to enable AI to continue writing. If you cannot make the acknowledgement, leave the feature off.

OpenAI

1. In your OpenAI API account, create a key and check model access and API billing. A ChatGPT subscription and API billing are separate.
2. In editmd, choose OpenAI under Provider.
3. Enter an exact Model ID available to your API account and a useful Connection name, such as “Writing”. Use the provider’s current model documentation rather than guessing a model name.
4. Enter the API key. Read the provider disclosure and select its acknowledgement.
5. Choose Save connection, then Test connection under the saved connection.
6. Check that the test reports “Structured output is available” before using the model with a document.

Remote keys are stored in macOS Keychain. A connection test uses fixed synthetic text, not your open document; it is still an API request and may incur provider charges.

Provider help: OpenAI API quickstart and ChatGPT versus API billing.

OpenRouter

1. Create an API key in your OpenRouter account and check your balance and account controls.
2. In editmd, choose OpenRouter — model gateway.
3. Enter your API key, then choose Load compatible models.
4. Use Search models and choose an exact Model ID from the list.
5. Add a Connection name, review the provider disclosure and select its acknowledgement.
6. Choose Save connection, then Test connection. A passed test enables document actions for the rest of the current app session; test again after relaunching.

editmd filters for compatible routes and requires specific privacy and structured-output controls. A model being advertised on the provider’s website does not guarantee it can be used through this connection.

To change an existing connection, use Switch model using saved key, choose the new model and test again. You do not need to disconnect merely to choose another model.

Provider help: OpenRouter quickstart and API keys.

Ollama on your Mac

Ollama is a separate installation. Ollama’s own documentation currently lists macOS 14 or later, on Apple silicon or, CPU-only, on Intel. Models need additional storage and memory; choose one that fits your Mac and check its own requirements.

1. Install and open Ollama using its official Mac guide.
2. Download a local model through Ollama using the steps below. Choose a locally running model, not a cloud-routed model. editmd does not download or manage models for you.
3. In editmd, choose Ollama — local on this Mac.
4. Choose Refresh installed models, then select the installed Model ID. No API key is needed.
5. Read the local-provider disclosure, acknowledge it and choose Save local connection.
6. Choose Test connection and confirm the result lists Ask as supported. The result is kept only for the current app session, so test again after relaunching. A model that loads successfully in Ollama may still fail editmd’s grounding checks.

The connection uses the fixed local address 127.0.0.1:11434. editmd checks for a reported remote route before sending document content. Keep Ollama running when using Ask.

Download a local model

1. In Ollama’s model library, choose a local model and size that fit your Mac. Copy its exact name and tag. Model size, memory needs and licensing differ.
2. Open Terminal on your Mac. Replace MODEL_NAME in the command below with that exact name and tag, then press Return. Wait for the download to finish.

ollama pull MODEL_NAME

3. Check that the model is installed:

ollama ls

4. Return to editmd and choose Refresh installed models. Select the matching entry, save the connection and run Test connection.

MODEL_NAME is a placeholder, not a model recommendation. If Terminal cannot find ollama, finish Ollama’s Mac installation and reopen Terminal. See the Ollama command reference. Downloading a model does not guarantee it will pass editmd’s grounded-Ask checks.

If the connection test fails

Check the provider, exact model ID, key access, billing or balance, and network connection. For Ollama, confirm the app is running and the model is installed locally. For OpenRouter, refresh the compatible catalog and test the chosen model again.

A “structured output unavailable” result means that model cannot provide the response format editmd needs. Choose another compatible model; changing a display name does not change capability.

When setup succeeds, continue with your first AI request.