YOUR AI. YOUR MACHINE. YOUR CONTROL.
Stop renting AI.
Start owning it.
No technical expertise needed.
Chat. Create. Put your agent to work.
Free AI that runs locally on your Mac.
No token bill. No cloud model required. Just get started.
MAXIMUS — THE OWNERSHIP MANIFESTO
The future of AI isn’t rented.
It’s owned by you.
Your ideas. Your work. Your next move. Not everything needs to pass through someone else’s cloud. Maximus offers another way: AI that works where the value is created. With you.
Six principles. A different relationship with AI.
Intelligence belongs with you.
On your machine. Close to your work. You choose what Maximus can read.
Share deliberately.
Work locally. Mask sensitive details when you need to take text elsewhere. News and web research use the internet.
Your agent works. You speak for yourself.
Let it read, monitor and prepare your work. You make the decisions and choose what to send.
Think more. Pay nothing per thought.
Local AI without token charges. From your first question to your next draft.
One less rented dependency.
Start with one task. Then a working day. Do more with the machine you already have.
You don’t need to be technical.
You have decisions to make, not local AI to study. Describe what you want done: explore an idea, make sense of a document, create a file or ask your agent to monitor something. You don’t need to learn about tokens, context windows or model settings to get started. Open Maximus. Start with your task.
Everything Maximus can do
The assistant responds when you ask. The agent follows your tasks while you do something else — while the app is running and the computer can work. You choose access; mail, calendar and other sources require your permission. Web features require internet.
The assistant
Transcribe a recording
Speech becomes text here on the computer — no audio leaves the machine
Read a document
PDF, Word, RTF, ODT and images — the text is extracted locally
Crunch a spreadsheet
Sum, average, median and extremes — calculated, not guessed
Write with real details
Names, ID numbers, amounts — they stay on this computer
Get a masked version
The same text with names and numbers swapped out, to take with you
Ask several questions at once
The question is split up, and you get one answer per part
Write a draft
Email, letter or statement — in its own block to copy
Reframe a text
Different tone, length, reader — you get the finished text
Make a text vaguer
Amounts, dates and unusual details are rewritten so they reveal less
Summarize a long document
The essentials first, with page references back
Compare two versions
What changed between two documents, and what it means
Look up something current
Amounts, rates and rules that change — with sources
Dig into a question
Several rounds on the web, with what each round turned up
Long conversations. Less to keep track of.
Older turns are automatically summarised when context fills up. Recent messages stay in full.
Gather conversations in a project
A new conversation can read what was said in the earlier ones
Track a deadline
MAXIMUS tells you when something is coming up — not afterwards
Work with an email
MAXIMUS reads — never writes, never sends
Lock a conversation
A code of its own on just this case
Share a conversation encrypted
A file and a password — no service in between
Export a decision brief
The whole case as a PDF — questions, answers, sources and receipts
Read the ledger
Every payload that has left the computer, in order
Read the calendar
What's on this week, without anything written back
Set a password to lock the disk
Without it, your cases lie in plain text on this computer
The agent
Let the agent go through the inbox
Every morning: what concerns you is raised, the rest set aside with reasons
Prepare this week's meetings
The agent reads the calendar and flags meetings missing material
Watch your reminders
What's due soon is raised before it's too late
Pick out messages waiting for a reply
iMessage and SMS, read on the computer — nothing is sent
Watch a folder
New files in Downloads are read and summarized for you
Talk to the agent
One conversation for everything it does: "what happened last night?", "pause everything until Monday"
Let the agent go through everything
What's closest, what's waiting for you, and three suggested tasks
Ask about your week
The answer is read from the calendar, inbox and reminders — not guessed
Tell me when someone emails
The agent looks when something new arrives, not just on a schedule
Watch a topic
No address needed — the agent finds the sources and shows them
Search for me in the background
Finds with image, price and link — and a notification when it's done
Remind me, book, make a draft
The agent suggests — a reminder, a meeting, an email draft — and does it after your yes
The agent has found things for you
What it has raised since you last looked, with reasons
Also: create editable Word documents and PowerPoint presentations, record meetings, dictate, get personal news and choose a local model or an optional cloud provider. Live dictation requires macOS 26; audio files and meeting transcription use Whisper. Optional external services may charge their own fees.
Matched to your Mac. Your choice of model.
You choose the task. Maximus suggests the size.
At setup, Maximus checks your memory and free disk space and suggests both a language model and a transcription model. It prioritises the largest tested language model that fits with headroom. It selects a model; it does not automatically resize the same model.
Which MacBooks?
MacBook Air and MacBook Pro with Apple Silicon (M-series), macOS 12 or later. 16 GB memory works; 24 GB or more is recommended for Gemma 4 12B. Intel Macs are not covered by the current Mac release. Smaller options exist for 8 GB Macs, but are not marked as tested in Maximus. Performance also depends on other workloads.
| Memory | Typical suggestion¹ | Tested in Maximus |
|---|---|---|
| 8 GB | Gemma 4 E2B · ~3.1 GiB | No |
| 16 GB | Gemma 4 E4B · ~4.8 GiB | Yes |
| 24 GB+ | Gemma 4 12B · ~6.5 GiB | Yes |
¹ Assumes enough free disk space. E4B and 12B are marked as tested in the app’s model catalogue. Testing does not guarantee every answer is correct. Model files download during first setup; this can take time.
Switch when you want
Choose your own and the model settings offer more local options: Qwen3, Llama, Mistral and larger Gemma models. They have their own memory requirements and licences, and are not marked as tested in the catalogue. You can also select your own compatible GGUF file. Cloud models are optional, use external services and may cost money.
One model to think. One to listen.
For Swedish transcription, KB-Whisper large is suggested with at least 16 GB memory and enough disk space. Whisper large-v3-turbo is suggested for other languages, less memory or limited disk space. Both are built-in model choices; the audio catalogue does not carry the same test labels as the language-model catalogue. Switch between them, choose your own compatible whisper.cpp file (.bin), or skip the audio model. Audio processing is local.
A TASTE OF MAXIMUS
News for you
See what matters.
News shaped around your work.
Recorded app · example data · waits shortened
Pick a feature. Get a feel for it before you download.
Run it. Own it. Be sovereign.