What the AI does
Write or paste anything into the AI Classifier, for example:
and it might come back as five separate tasks:
- Send the Q3 numbers to Sara, due Friday, in Do First: someone is waiting and the deadline is close.
- Reply to the landlord about the lease, due today, in Do First.
- Book a car service in Schedule: it matters, but nothing makes it urgent today.
- Dentist appointment, next month, in Schedule.
- Clean up the photos folder in Eliminate, the place for “maybe”.
The exact quadrants depend on the model, and you check every task before it is added. Here is what happens to your text:
- It is split into tasks. Two actions become two tasks, and filler and greetings are ignored.
- Each task gets two answers: is it important, and is it urgent? Together they choose the quadrant. Health, money, legal matters, your core work or studies, family and long-term goals count as important; everyday chores and errands do not, unless you say why they matter. Today, tomorrow, a close deadline or someone waiting make a task urgent.
- The app, not the AI, reads the dates and times in your words, so “next Friday at 3” becomes a real date by the same rules every time. See Dates, times and reminders.
- Nothing invented gets in. Every task must come from words you actually wrote. A task the model made up is not added.
- You review the result, change anything, and only then are the tasks added.
Choosing how AI runs
Open Settings › AI and pick one under Sort tasks with. The welcome tour offers the same choice the first time you open the app.
| Choice | Cost | Where your text goes | Good for |
|---|---|---|---|
| No AI | Free | Nowhere. It never leaves your computer. | Sorting tasks yourself |
| Eisenhower Pro | $7 a month | To the Eisenhower AI server | Nothing to install and no key to paste |
| Your own API key | What the provider charges you, usually a fraction of a cent per list | Directly to Gemini, OpenAI, Anthropic or Groq | Fast results on any computer |
| Local AI (LM Studio) | Free | To LM Studio, on your own computer | Privacy, and a computer with plenty of memory |
You can switch at any time; nothing about your tasks changes when you do.
Eisenhower Pro
Eisenhower Pro runs the classifier on the Eisenhower AI server, so there is nothing to install and no API key to manage. It is a paid account, shared with the Eisenhower AI iPhone app: one account covers your computers and your iPhone.
- Open Settings › Account and choose Set Up Pro, or choose Eisenhower Pro in the welcome tour.
- Sign in with your email address and password, or create an account. A new account, or a password reset, is confirmed with the 6-digit code from the email, typed into the app.
- In Settings › AI, choose Eisenhower Pro under Sort tasks with.
Settings › Account shows who is signed in, how many computers are signed in (Pro allows three), and the subscription with its usage. Sign Out removes this computer from your account. If an account has no Pro subscription, the AI panel says so, and Local AI or your own key still work.
Your own API key
With your own key, the app sends your text straight to the AI company you pick, and that company bills you for what you use. Sorting a list is a small request, so even daily use usually costs very little; some providers also have a free tier. Check each provider's current prices on its site.
| Provider | Model the app uses | Get a key | Key starts with |
|---|---|---|---|
| Gemini (Google) | gemini-3.5-flash | Google AI Studio | AIza |
| OpenAI | gpt-5.4-mini | OpenAI Platform | sk- |
| Anthropic | claude-haiku-4-5 | Anthropic Console | sk-ant- |
| Groq | llama-3.1-8b-instant | GroqCloud | gsk_ |
Adding a key
- Create a key on the provider's site, and copy it. Some providers need a payment method or credit on the account before the key works.
- In Eisenhower AI, open Settings › AI, choose Your own API key, and pick the Provider.
- Paste the key into API key and press Save key.
- Press Test AI. It sends one short sentence, “Reply to the vendor by Friday”, and tells you which model answered and how long it took.
Saving checks only that the key looks right. A key that is mistyped, revoked or out of credit shows up when you test it or sort a list.
Where the key is kept
The key goes straight into your system's secure storage: the Keychain on a Mac, Credential Manager on Windows. The app's window only ever sees a masked copy, and the key is sent only to its own provider. Delete key removes one; Remove all keys under Saved keys removes them all.
What the provider receives
Local AI with LM Studio
LM Studio is a free app that runs AI models on your own computer. Eisenhower AI talks to it over a connection that never leaves your machine, so your text stays private and nothing costs anything per use.
What it needs
Setting it up
- Download and install LM Studio, then open it once. The first launch sets up the
lmscommand that Eisenhower AI uses to start the server and load the model. - Download a model in LM Studio by searching for its name, or from a terminal:
lms get google/gemma-4-26b-a4b lms server start --port 1234 - In Eisenhower AI, open Settings › AI and choose Local AI (LM Studio).
- Check Local model name matches the model you downloaded, exactly as LM Studio writes it.
- Leave Manage LM Studio automatically on, and press Test AI. The first test can take up to two minutes while the model loads.
Which model
| Model | In the app | Notes |
|---|---|---|
google/gemma-4-26b-a4b | The default: “Finds more tasks, same speed” | A 26 billion parameter model and a download of several gigabytes. Needs the most memory. |
qwen/qwen3.5-9b | “Smaller, lighter on memory” | A much smaller download that sorts a little less accurately. |
What “Manage LM Studio automatically” does
With it on, each time you sort a list the app:
- starts LM Studio's server if it is not running;
- loads your model if it is not loaded, and lets LM Studio unload it after five idle minutes.
Turn it off if you prefer to run LM Studio yourself. The app then only sends requests to the LM Studio server URL, which is http://127.0.0.1:1234/v1 unless you change it. The address must be on this computer; the app refuses any other, and never sends Local AI requests through a proxy.
If the test fails
A checklist, What Local AI still needs, appears under the test, marking each step Ready or Needs attention:
- LM Studio is installed. If it is not found, install it and open it once, so the
lmscommand is set up. - Its local server is running. Start the server in LM Studio's developer panel. If it shows a different port, put that port in the server URL.
- The model is downloaded. Download it in LM Studio, or copy the
lms getcommand shown and run it in a terminal.
Fix the step marked, then press Check again.
Sorting a list
- Press AI in the title bar to open the AI Classifier. The first time, a short How it works card explains it, and Try an example fills in a rough paragraph to sort.
- Paste or type into the box, or press Mic and speak.
- Press Classify tasks. With Pro or your own key it usually takes a few seconds; with Local AI, longer.
- In Review AI Tasks, change titles, quadrants, dates and times, tick Remind for a reminder at that date and time, or remove a task. Then press the Add button, which counts the tasks.
The tasks are added to the matrix you have open. The whole batch is one change, so Undo takes it all back.
Settings that change this
Under When AI sorts in Settings › AI:
- Review AI tasks before adding. Turn it off to add tasks straight away. Review still opens if the AI could not place a task, because every task needs a quadrant.
- Show coaching after sorting: a short note on what was added and how the quadrants are balanced.
Limits
Up to 50 lines at a time, which is a long brain dump. Split anything bigger into two. Dates are read in English.
Dates, times and reminders
The app reads dates and times from your words, so you rarely have to set them by hand. Some examples, as if written on a Wednesday afternoon:
| You write | The task gets |
|---|---|
| “call the bank tomorrow at 9” | Thursday, 09:00 |
| “send the report by Friday” | Friday |
| “next Monday at 8 am” | The coming Monday, 08:00 |
| “tonight” | Today, 20:00 |
| “in 2 hours” | Today, at that time, to the next 5 minutes |
| “shortly”, “in a bit” | Today, no time |
- When one task names two days, as in “the demo is Friday at 11, prepare it tomorrow”, the task gets its own day (tomorrow), and the 11 stays with the demo.
- When a list holds several tasks, each keeps its own date: “finish the report by Friday, call Mom this weekend, renew my passport before December” gives three tasks with three dates.
Reminders
A task gets a reminder when it has a date and a time and you ask for one: “remind me”, “notify” or “alarm”. “Remind me to call grandpa in 2 hours” reminds you in two hours. You can always tick or untick Remind in review.
Voice input
Press Mic in the AI Classifier, say your tasks, and press Stop. What you said appears in the box, ready to sort. Set the language you speak in Settings › AI › Voice language.
Mac
macOS's own speech recognition turns your speech into text, and may send the audio to Apple to do so. The first time, macOS asks whether the app may use the microphone and speech recognition; change either answer later in System Settings › Privacy & Security.
Windows
Speech is turned into text on your computer, by a speech model the app downloads once (190 MB). Your voice is never stored or sent anywhere. The microphone records only between Mic and Stop, for at most two minutes.
The model needs a processor with AVX2, which most made since 2013 have; if yours cannot run it, the app says so before downloading. Remove speech model in Settings › AI deletes it.
When something goes wrong
If sorting fails, your text stays in the box, so you can try again.
| Message | What to do |
|---|---|
| “Turn on AI in Settings to sort a list of tasks.” | Choose Pro, your own key or Local AI under Settings › AI. |
| “Save an API key for … in Settings to sort tasks with it.” | Paste and save a key for the provider you picked. |
| “Signed in, no Pro subscription” | The account has no Pro subscription. Use your own key or Local AI instead. |
| “… classification request failed (401)” or “(403)” | The key is wrong or revoked, or the account has no access. Create a new key, and check the account has credit. |
| “… classification request failed (429)” | Too many requests, or the quota is used up. Wait a minute, or check your plan with the provider. |
| “… classification request failed (500)” and similar | The provider is having trouble. Try again in a few minutes. |
| “LM Studio CLI (lms) was not found.” | Install LM Studio and open it once. |
| “Cannot reach Local AI at …” | Start LM Studio's server, or fix the server URL. |
| “LM Studio is already running on port …, but Eisenhower AI is configured for port 1234.” | Change LM Studio to port 1234, or put its port in the server URL and turn off Manage LM Studio automatically. |
| “The selected local model … is not installed in LM Studio.” | Download the model, or correct the Local model name. |
| “Sort at most 50 tasks at once.” | Split the list in two. |
| “AI response did not contain JSON.” and similar | The model gave an answer the app could not read. Try again; with Local AI, try the default model. |
The first Local AI request after launch loads the model and can take about two minutes, and the AI Classifier says so; give it time before cancelling. More fixes are in Help and FAQ.