Think inside your AI world.
Why does ChatGPT forget my preferences?
Usually one of three mechanics: the preference never made it into memory, it was there and got compressed away, or it was phrased as a passing remark the distiller didn't keep. All three have fixes.
ChatGPT's memory is real, but it is selective by design: a background process distils your chats into short fact-like entries, and it keeps what looks durable. A preference mentioned once mid-task often doesn't clear that bar. Within a single long chat, a separate mechanic bites: the context compresses, and instructions from early in the conversation fade. The fixes are different for each.
Making a preference actually stick
State it as a standing instruction, not an aside — 'always write dates as 12 June 2026' lands better than a one-off correction. Check Settings → Personalisation → Memory to see what was actually kept; you can delete wrong entries and add explicit ones. For preferences that must hold everywhere, custom instructions outrank memory: they are injected every chat, verbatim, no distiller in the loop.
If a preference holds early in a chat and decays late, that is compression, not memory — long conversations get summarised under the hood, and summaries keep gist over instruction. Restating the rule near the work is the practical workaround.
Why some things never stick
The memory layer is fact-shaped on purpose: small, stable, low-risk entries. 'Prefers British spelling' fits. 'We agreed the newsletter leads with the customer story, because the last three product-led issues underperformed' mostly doesn't — it is a decision, with reasoning, that could be revisited and superseded. A store built for facts has nowhere to put the reasoning or the status, so the distiller either drops it or keeps a flattened version that reads as trivia.
That flattening is why the you-shaped facts persist while the work-shaped calls dissolve — briefed on who you are, blank on what you settled.
Where the work-shaped calls can live
Unl holds exactly that unit: the call you settled, the why you settled it, and whether it still stands. Connected to ChatGPT over MCP — and tuned first on Claude and Claude Code, where it is fullest — it serves the standing decision into the session at the moment it is relevant, without you re-stating it. Preferences belong in custom instructions; settled judgement belongs somewhere that tracks supersession.
Ask which kind of thing keeps being forgotten. If it is a fact about you, tune the memory. If it is a call you made, it needs a home that holds the why.
ChatGPT keeps you-shaped facts; the calls you make about the work need a home that carries reasoning and status — and that difference, not a memory fault, is usually why your 'preferences' keep vanishing.
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
How do I see what ChatGPT has actually remembered?
Settings → Personalisation → Memory lists every stored entry. You can delete anything wrong and explicitly tell it things to keep. If an important preference is missing, say it as a standing rule in chat, or move it into custom instructions, which apply to every conversation without depending on the distiller.
Why does it hold a preference early in a chat and lose it later?
Long chats get compressed so the model can keep working within its context limit, and compression keeps the gist of the discussion over standing instructions. That is a within-session mechanic, separate from cross-session memory. Restating the rule close to where the work happens is the reliable workaround.
Can Unl work with ChatGPT?
Yes — Unl connects over MCP, and ChatGPT can add MCP connectors. It is built for and tuned on Claude and Claude Code first, and fullest there; what it carries into any surface is the same: your settled decisions with their reasoning, served when relevant.
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.