Setting up Inkling, start to finish.

Before you start

Inkling does the AI work on your own machine, so unlike a web app it has real hardware to ask about. Here is the honest list.

macOS 13 or later

Ventura and up. Inkling uses Apple's modern audio-capture API, which older macOS versions do not have.

Apple Silicon recommended

M1 or newer. Intel Macs run it, but the on-device AI is noticeably slower.

About 3 GB of free disk

For the two AI models Inkling downloads on first run. They are stored once and reused.

8 GB of memory, 16 GB is comfortable

The default note-writing model wants roughly 3.5 GB free while it works. There is a smaller one if your Mac is tight.

Internet for the first run only

To fetch the models. After that Inkling records, transcribes, and writes notes with no connection at all.

No account

There is nothing to sign up for, nothing to activate, and no license server to phone.

Install & permissions

Downloads open at launch. Installing is the ordinary Mac routine, with no installer to click through, and no account to create.

1. 1Download the Inkling .dmg and open it.
2. 2Drag Inkling into your Applications folder.
3. 3First launch: macOS may warn it cannot verify the app. Right-click Inkling, choose Open, and confirm; every launch after is a normal double-click.
4. 4Grant Microphone and System audio when asked (both required), and Calendar if you want it (optional).

First run: the two downloads

Inkling has no cloud to lean on, so the intelligence has to live on your Mac. On first launch it walks you through four screens, and two of them are downloads. Together they come to roughly 3 GB and it happens once.

Downloads run in the background and resume if they are interrupted, so you can cancel one and come back to it. Once both are on disk, Inkling never needs the network to record, transcribe, or write again.

Changing the transcription engine

Live in Settings → Transcription. The two local engines are the point of the product and cover almost everyone.

• Parakeet: recommended

  Fast and accurate on Apple Silicon, and the reason transcripts appear while you are still talking. Runs entirely offline.

• Whisper: the alternative

  Several sizes, from a 74 MB model up to a 3 GB one. Bigger is more accurate and slower; the default balance is Large v3 Turbo at about 1.5 GB. Also fully offline.

Who writes your notes

Open Settings → Summary and you get three choices under Connect your AI. The first is already set up and working. The other two are there for when you want more.

Using Ollama

Ollama is a free app that runs open-source AI models on your own Mac. Pair it with Inkling when you want a bigger, better model than the built-in one without giving up the thing that makes Inkling Inkling: nothing you feed an Ollama model leaves your machine.

1. Install Ollama

Download it from ollama.com/download, drag it to Applications like any other app, and open it once. It puts a small icon in your menu bar and quietly runs models for anything on your Mac that asks. If you use Homebrew, this works instead:

2. Pull a model

A model is a separate download from Ollama itself. Open Terminal and pull one. This is a good starting point and a clear step up from the built-in model:

If your Mac has 32 GB of memory or more, a larger model writes noticeably better notes:

To see what you already have, or to confirm Ollama is running at all:

Any chat model in Ollama's library will work. Bigger models write better and run slower, so match it to your Mac rather than reaching for the largest one on the list.

You can also skip the terminal entirely: if Inkling sees Ollama installed with no models, it offers to download a small one for you.

3. Point Inkling at it

1. 1Open Settings → Summary in Inkling.
2. 2Under Connect your AI, choose Your Ollama. Inkling checks whether Ollama is running and tells you what it found.
3. 3Click Fetch Models. Everything you have pulled appears in the list below, with its size.
4. 4Click the model you want, then save.

Leave the endpoint field empty unless you know you need it. Inkling talks to Ollama at http://localhost:11434 on your own machine by default. If you run Ollama on a different computer on your network, put that address in and click Fetch Models again.

One thing worth knowing: Ollama has to be running when Inkling writes a note, not only when you set it up. It launches with your Mac by default, so in practice this only bites if you have quit it deliberately.

Using your own API key

A frontier model writes the best notes of any option here. Inkling has no subscription and no server in the middle, so the way to use one is with your own key: you sign up with the provider, you pay them directly for what you use, and Inkling talks to them on your behalf.

Where to get a key

Copy the key when it is first shown to you. Every one of these providers shows it exactly once and only stores a hash afterwards, so if you lose it you make a new one.

Add it to Inkling

1. 1Open Settings → Summary.
2. 2Under Connect your AI, choose Your API key.
3. 3Pick the provider you got the key from.
4. 4Paste the key into the API key field. It is masked as you type; the eye icon reveals it and the padlock locks it against accidental edits.
5. 5Click Test connection. Inkling makes one tiny real call and tells you whether it worked, rather than failing quietly on your next meeting.
6. 6Choose a model. The list fills in with the models your key can actually reach, so you are not guessing at names.

Custom and self-hosted servers

If you already run a model server, Inkling will talk to it. Anything that speaks the OpenAI chat API works: LM Studio, llama.cpp's server, vLLM, a LiteLLM proxy, or a private endpoint at your company.

1. 1Open Settings → Summary → Your API key, and choose Custom Server (OpenAI) as the provider.
2. 2Enter the base URL, including the version path, for example http://localhost:8000/v1.
3. 3Type the model name exactly as your server exposes it. There is no list to pick from, because only your server knows what it is serving.
4. 4Add a key if your server requires one, or leave it empty if it does not.
5. 5Click Test Connection before you rely on it.

Max tokens, temperature, and top-p are all adjustable here too, for the case where you are tuning a local model and know what you want from it.

When something is not working

Still stuck? Tell us what happened and a person will read it.