The argument
The wrong-shaped unit
AI memory works mechanically and still feels wrong. The reason isn't the model, or the engineering — it's the unit. Memory ships an agent-shaped unit into a human environment. Unl holds the unit you actually think in.
Chat is a human environment
You don't recall facts. You recall what you settled, and why.
Think about how you actually carry a project in your head. Not transcripts — you couldn't replay last Tuesday's conversation if you tried. Not a pile of facts about yourself. What you hold are settled points: the calls you made, each with the reasoning attached — and when your thinking moves, the old call doesn't linger, it's superseded.
That's the native unit of human thought in a working relationship. Any context layer built for a place where the user is a person should be built out of that unit.
What AI memory actually ships
An agent-shaped unit, carried into a place where the user is you.
The memory field builds transcripts, embeddings, fact-triples, belief states. And to be fair: agents genuinely need those. An autonomous agent maintaining its own continuity needs a machine-queryable record of state — that unit is right for that job.
The trouble starts when the same unit is ported into the human environment. It works mechanically — things are stored, things come back. It fails relationally: there's no gesture where you said keep this, no moment where you ratified what it holds, and no why you can recognise, stand behind, or override. It remembers about you; it doesn't hold anything with you.
The human-shaped unit
A decision, with its reasoning, that you actually settled.
Unl holds that unit and nothing else: the calls you've made, each with the why attached, superseded when your thinking moves — and served back as unprompted relevant context, arriving the moment it bears on the work rather than waiting to be asked for.
You feel the difference in the first session: a cold chat starts from nothing and re-argues everything; a briefed one starts from what you settled and moves. Tuned for Claude, Claude Code, ChatGPT & Cursor at launch, extending across the AI ecosystem. Connects anywhere MCP does.
And the twist — the model prefers your unit too. A reasoning model works better from reasoning than from answers alone (the basis of chain-of-thought prompting). Hand it a bare conclusion and it re-derives its own: a no without a why is a yes to a reasoning agent. The human-shaped unit is also the machine-optimal fuel. One shape, both audiences, no compromise.
Questions people ask
What's the difference between Unl and AI memory?
They hold different units. AI memory stores what came up — transcripts, extracted facts, embeddings — a unit built for agent continuity. Unl holds what stands: the decisions you settled, each with its reasoning, superseded when your thinking moves. Memory recalls; Unl carries your settled reasoning into the model's context the moment it bears on the work.
Is AI memory bad, then?
No — it's the right unit for a different job. Autonomous agents genuinely need machine-queryable state: logs, embeddings, belief states. The mismatch only appears when that agent-shaped unit is carried into a human environment, where there's no gesture, no ratification, and no why a person can recognise or override.
Why does the reasoning matter to the model, not just to me?
Because reasoning models reason better from reasoning — that's the basis of chain-of-thought prompting (Wei et al., 2022). A bare conclusion gives the model nothing to build from, so it re-derives its own answer and can re-open what you settled. A decision with its why binds.
How does Unl get the right decision in front of the model?
As unprompted relevant context: when something you settled bears on the current turn, Unl serves it — the decision with its reasoning — without you having to remember it exists. You stay in command of what's held: it's saved on your word and superseded on your word.
What this is
Think inside your AI world — you stay in command
Unlimitless (Unl to friends) holds what you've settled, reads what your tools are showing, and catches what's changed out in the world — and hands your AI whatever bears on the work, the moment it's needed, without you asking. The right thing, in front of the model, unprompted, with you in command of the call. So you keep moving toward what you set out to build, on top of everything you've already decided.
It plugs into Claude, Claude Code, ChatGPT and Cursor as an MCP connector. Quick to connect, in a couple of steps.
Unlimitless is open now to invited Alpha. Apply for the Beta waitlist to come in ahead of the full launch:
Alpha is invite-only · free at launch.