Answers
I keep a spreadsheet of facts to paste into every AI. Is there a better way?
Yes, and the spreadsheet is the right instinct in the wrong place. Keeping a short, chosen list of facts and decisions is exactly what makes an assistant useful — the part that breaks is that you are the delivery mechanism. Arroway keeps that same list on a server your AI tools connect to, so each assistant reads it at the start of the work instead of waiting to be given it. You keep the part that was always yours, which is deciding what belongs in the list, and stop doing the part no person should be doing, which is carrying it from tool to tool.
Last updated September 3, 2026
The list is not the problem. Being the transport is
Nothing about a hand-kept list is wrong. It is short, a person chose every line, and it says what is actually true about how you work — which is more than most automatically gathered context manages. The cost sits everywhere else. It only reaches an assistant when you remember to paste it, it reaches each one in a slightly different state, and the day it grows past what fits comfortably in one message you start choosing which parts of your own memory to leave out. That choice gets made under time pressure, in the middle of a task, and it is never revisited.
Same curation, served instead of pasted
Each row of that spreadsheet becomes an entry carrying the two things a cell has nowhere to put: who decided it, and what would end it. Your assistant reads the set when it starts working and writes new entries back as they come up, so the list grows alongside the work rather than during a tidy-up you keep postponing. What it does not do is grow behind your back — a new entry is a proposal until a person approves it, and approving is a few seconds in a review, not a project.
One list, every tool, no second version
The reason to move the list off your own machine is not storage. It is that a file on one computer has to be copied to be used, and every copy is a version that will eventually disagree with the others. Here there is one set, and Claude, ChatGPT, Codex or anything else that speaks the same connection standard reads that set. A second machine, a new tool, or handing part of the work to someone else stops meaning "export the sheet again".
What it looks like in practice
Someone keeps 292 rows spread across three surfaces: a spreadsheet, a notes app, and a long preamble they re-paste at the start of each session. Perhaps forty of those rows are dead — a library version that moved twice, a client rule that was reversed, a preference abandoned months ago. They know roughly which ones, so they hedge: paste the list, then correct the assistant in the chat, twice a day, from memory. The rows were never the problem. Every one of them was worth writing down. What a spreadsheet could not do was say which lines were still in force, get itself read without being asked, or let a dead line be retired in the same place it was written.
Questions people ask about this
- Is this not a spreadsheet with extra steps?
- The difference is who reads it, and when. A spreadsheet is read by you and then re-typed for a machine; this is read by the machine at the start of the work, with no gesture from you. And each entry carries what a cell has nowhere to hold — the person who sanctioned it, and the condition that ends it — which is what lets a dead line be found and retired instead of quietly outliving its usefulness.
- Can I bring the spreadsheet I already have?
- Yes, and that is the normal way to start. An import takes the rows you already trust, and the review afterwards is usually the first time anyone reads the whole list in one sitting. Rows that turn out to be dead do not need deleting one by one — you simply leave them out of the import.
- What stops this becoming the same mess, just bigger?
- Two things a spreadsheet has no room for. Each entry says what would make it wrong, so a rule tied to a version, a client or a quarter carries its own expiry instead of waiting to be noticed. And the product surfaces what looks stale, redundant or unconfirmed, so you are shown the candidates rather than having to go looking. A person still decides what goes: judging what is no longer true is judgement, not cleanup.
Where this is verifiable
Product documentation on this site (How it works, Install), the import and export tools that ship with the connection, and the sanctioned product spec in the repository. Everything described here is behaviour the tools apply today, not roadmap.
https://www.arroway.app/en/answers/tracking-ai-facts-without-a-spreadsheet