MYNDAGENT LOOPTHE COREITS OWN COMPUTERLAST 20,000 TOKENSOVER A BRIEFYOUR MESSAGE, MID-TASKUP TO 3 AT ONCEONE EXACT ACTION, ONCEHOURLY 7 AM–11 PMSOURCE-CHECKEDUPKEEP 3:30 AM123456123456

Every world deserves its own intelligence.

A life is a world, so is a family or a company. Axel Labs is building AI that understands each one in depth. We’re starting with Axel, a personal AI agent.

01

AI is getting smarter every month. It still doesn’t know you. We’re building personal intelligence.

Life and work are chaotic, with a thousand things to keep track of and make sense of together. Mynd is the system we built to capture those details and hold them for you, so AI can understand you and your life in depth.

02

Mynd connects it all into one picture of you.

Each new detail gets linked to what came before: people, plans, money, work. Over time that becomes a picture of what matters to you and how it all fits together, and every fact keeps its source. How Mynd works

03

Axel runs on a harness built for work that doesn’t stop.

An agent is only as good as the system around the model. Axel’s harness gives him a persistent computer, one continuous working context, loops that run when no one is asking, and authority enforced outside the model. Mynd supplies the context each step needs.

  • One continuous contextEach run reopens the same session where the last one stopped, not a summary of it. While no one waits, older turns fold into a brief; the latest 20,000 tokens stay word for word.
  • Interruptible turnsA message sent mid-task is answered in the same turn and redirects the work already running, while slow steps keep executing. A run that hits one reply’s output limit picks up where it stopped.
  • Background and proactive loopsAn hourly look from 7 AM to 11 PM reviews mail, calendars, accounts and open work against the recent conversation, with a deeper pass each evening and memory upkeep at 3:30 AM. A second model checks every finding against its sources before you see it.
  • Self-review and learningEach night the agent reads back over its own runs, keeps the method that worked and records defects in its own tools. Steps repeated by hand become a routine that is tested before it is used.
  • Parallel sub-agentsUp to three helper agents run at once beside the main conversation, each with its own browser pages, and hand back one result.
  • Authority outside the modelStanding permissions live in the harness’s core, outside the model, and approvals are bound to a hash of the exact action and used once. Saved passwords are typed into the page by the core and are not shown to the model.

04

We’re building the whole stack, from memory to machine.

We’re developing every layer ourselves: Axel and custom apps on top, the harness and Mynd beneath them, then our own models and the inference that runs them, down to the Axel pod. Until our models are ready, Axel runs on strong models from established providers. The goal is AI that understands your life well enough to feel like personal superintelligence.The next frontier lab, made in Canada.

What we build

AI that learns your world, for people and for organizations.

Axel on iPhone, working through an allergy form step by step (opens in a new tab)

Invite-only betaFor people

Axel

Our first product is Axel, a personal AI agent with his own always-on computer. He remembers what matters through Mynd, works in your accounts with your OK and keeps working while you’re away. Money first, then the rest of life.

Visit axelagent.io (opens in a new tab)
The Axel pod: a small computer, its glass dome holding Axel’s world

In developmentInfrastructure

Our own models and the Axel pod

We’re developing our own models and the inference that runs them, and in time the Axel pod, a machine built around the model and Mynd.

Axel, in practice

Tell him what you need, the way you’d tell a friend.

He works it through on his own computer, in your accounts and with your OK, then tells you where it landed. There’s nothing to set up first and no prompts to learn.

  • Talks with you in the thread, or on a call you can take anywhere.
  • Works on his own computer while you watch, or take over.
  • Keeps your money in view: what came in, what went out, what’s coming up.
  • Finds the card, the code or the page you need, when you need it.
  • Brings you only what needs you, on your lock screen.
  • Works only for you. You decide what he can reach, and can take it back.
Axel on the web at dusk: asked why eating out is so high this month, he answers with a Money card showing October’s eating out day by day and where it went, and offers a plan
Axel on iPhone filling in an allergy form: he read the daycare’s email, recalled a photo, checked the calendar and scheduled a pickup
Axel on iPhone explaining why eating out ran $286 over this month, restaurant by restaurant
Axel shopping on his own computer: the right size applied, the best price found, the order ready to place
Axel remembering a wish from June: the scarf she loved, back in blue, wrapped and arriving Friday

For organizations

A status meeting becomes a question.

The state of every project lives with the people doing the work. Give each of them an agent, connect the agents through one Mynd, and the organization can answer for itself, showing where each part came from and where two parts disagree.

You ask

“Are we ready to launch the new plan on the 18th?”

  1. Engineering lead’s agent

    Billing changes are live. Two payment fixes are in review.

  2. Legal counsel’s agent

    New terms are drafted, not yet signed.

  3. Finance lead’s agent

    Price approved at $49 a month.

  4. Marketing lead’s agent

    Launch page and email are ready. The page shows $39 a month.

The organization’s answer

Not yet. Legal hasn’t signed the new terms, and the launch page shows $39 against the approved $49. Engineering is on track.

4 sources1 conflict
See how it works

Where we’re focused

We start with the memory, and work toward the machine.

  1. 01

    Mynd

    A memory system built for a whole life. Every detail keeps its source and links into situations and one big picture, recalled by exact text, keywords and meaning at once. You can see, correct or remove any of it.

  2. 02

    Our own agent harness

    Custom harness code designed together with Mynd and, in time, with our own models, so memory, harness and model work as one system instead of being wired together afterward. It already gives each agent a persistent computer, one continuous context and authority enforced outside the model.

  3. 03

    Our own models

    In development: agent models adapted from strong open checkpoints and, with consent, from how real tasks get done, served by our own inference runtime on rented GPUs first.

  4. 04

    The Axel pod

    A machine for the whole intelligence, with compute placed beside the model’s memory so weights stay put and only activations move. It starts with our own expert engine; custom hardware and, in time, our own chips follow.

The master plan

We intend to think as big as possible, and push our own limits, the company’s limits and the field’s limits.

  1. 01

    An Axel people rely on.

    An AI that holds a deep picture of your life through Mynd and gets real work done, starting with money.

  2. 02

    A business that can support it.

    Custom intelligence for organizations, and people inviting the people they care about.

  3. 03

    Our own intelligence.

    Models of our own, tuned with permission on how real work gets done.

  4. 04

    On hardware you control.

    The Axel pod is designed to run all of Axel, model and memory included, on one machine.

The north star is personal superintelligence: an intelligence that knows your whole life, and can reason across all of it while helping with the task in front of you.

For yourselfJoin the Axel betaInvite-only for now. Ask for an invite at axelagent.io. (opens in a new tab)For your organizationStart with a BlueprintAbout three weeks, at a fixed fee: we map how your organization works and show one workflow running in your own terms.