Think inside your AI world.
Do AI memory features store decisions, or just facts?
Facts, overwhelmingly — by design, and reasonably so. The interesting question is what a decision would need from a store that a fact never does.
Current memory features — ChatGPT memory, Claude's memory, most agent memory layers — distil conversations into compact, stable entries: preferences, biographical facts, project names, tool choices. An Indie Hackers thread put the gap in one line that has travelled widely since: most memory features store facts, not reasoning or decisions — 'user prefers dark mode' is useless next to 'we decided to skip auth and use magic links because our users are devs'. That is a community observation, and it holds up under inspection.
Why stores skew fact-shaped
A fact is a safe unit: small, stable, cheap to keep, embarrassing only if wrong. Distilling 'uses PostgreSQL' out of a chat is low-risk. A decision is a riskier unit — it was made against alternatives, for reasons, and it can be overturned next month. Storing it properly means storing the why and tracking whether it still stands. Storing it as if it were a fact strips exactly the parts that made it worth keeping.
So the distillers, sensibly, keep what fits their shape. Nothing is malfunctioning. The unit is just smaller than the work.
What a decision needs that a fact doesn't
Three things. Reasoning: 'magic links, because our users are devs' will bend differently under pressure than the bare conclusion. Status: decisions get superseded, and a store with no supersession serves dead calls as confidently as live ones. And arrival: a decision is worth holding only if it shows up when the work touches it — sat in a store it is archaeology, served in the session it is standing.
Fact stores have none of these slots, which is why a model can know a great deal about you and still reopen everything you settled.
The layer built decision-shaped
Unl keeps that unit natively: each decision with the reasoning it was made on, retired cleanly when you change course, and served into Claude, Claude Code or any MCP surface at the moment it bears on the work. It is not a better fact store — facts can stay where they are — it is the missing store for the other kind of context.
Memory features answer 'who is this user?'; a decision store answers 'what stands, and why?' — and it is the second question your sessions keep paying for.
Reads through Unl arrive with measured context — in the presence of the decisions you’ve already settled. The reach lane is live: one box, paste anything. If it speaks MCP, Unl can reach it. Readings arrive unprompted, the data beside the criterion; Unl is a courier, not a warehouse, and keeps only your keys and the frame.
Read further
Questions people ask
Where is the facts-not-decisions observation from?
An Indie Hackers discussion on handling AI forgetting between sessions — a community observation, credited as such, and quoted here because it names the distinction more cleanly than most product copy: the dark-mode preference versus the magic-links decision with its reasoning attached.
Could memory features just start storing decisions?
They would need supersession, reasoning capture and relevance-timed serving — at which point they have become a different product with a different trust model, not a bigger memory. The human also has to stay the author of what counts as settled, which is a workflow question as much as a storage one.
How does Unl decide a decision is worth keeping?
It doesn't — you do. Nothing is captured silently: you settle something and choose to keep it, with its why. Unl holds it, tracks whether it still stands, and serves it when relevant. The judgement about what is settled stays yours.
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.