The thesis
You recall reasoning, not facts
A century of memory research agrees on the headline: human memory is reconstructive. You keep the gist and the why; the details get rebuilt on demand. So why is every AI memory tool built to keep the details?
How you actually remember
Gist and why kept; details rebuilt.
You don't store last month's meetings verbatim — you store what they settled and why it mattered, and you reconstruct the rest when asked. This isn't a flaw your tools should compensate for; it's the design. Reasons compress better than records, and they're what you need to act.
A context layer built for a person should hold what a person holds: settled points, each with the reasoning attached, retired when thinking moves on.
Why the tools were built the other way
Memory tools keep details because they were built for agents first.
Verbatim logs, embeddings, fact-triples, belief states — these are the units of agent continuity, and for agents they're genuinely right: a machine resuming its own work needs machine-queryable state. The field built its unit for its telos.
Carried into the human environment, the same unit works mechanically and fails relationally: nothing in a log is yours — no gesture put it there, no ratification stands behind it, no why is attached that you could recognise or override. Unl is the unit built the other way round: decisions ratified on your word, each with its reasoning, held up as unprompted relevant context when they bear. Tuned for Claude, Claude Code, ChatGPT & Cursor at launch, extending across the AI ecosystem. Connects anywhere MCP does.
The double whammy: the unit that mirrors how you think is the unit a model reasons best from. Chain-of-thought is the proof in miniature — models reason better from reasoning. Build the layer for the human and the machine gets its optimal fuel free.
Questions people ask
How should AI context work for a person?
The way human memory works: keep the settled points with their reasoning — the gist and the why — and let details be reconstructed when needed. Serve the relevant settled point unprompted when it bears on the work, and retire it when the person's thinking moves.
Why do AI memory tools store transcripts and embeddings instead?
Because they were built for agent continuity first — and for agents, machine-queryable state is the right call. The mismatch appears when the agent-shaped unit is carried into a human environment without the gesture, ratification, or why that human recall runs on.
Is keeping less actually better?
For this job, yes. Reasons compress the record without losing what carried it: one settled decision with its why replaces every conversation that led there — and unlike a pile of details, it can't be mistaken for an open question.
Do models really prefer this unit?
Reasoning models work measurably better from reasoning than from bare answers — the finding behind chain-of-thought prompting (Wei et al., 2022). The human-shaped unit is also the machine-optimal one; nothing about serving people better costs the model anything.
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.