Answers
Arroway vs writing it in CLAUDE.md: what is the difference?
CLAUDE.md is a file in a repository: someone writes it, someone keeps it up to date, and the coding assistant working in that repository loads it. Arroway is a shared memory that every AI tool connected to a project reads before it acts — Claude, ChatGPT, Codex and other tools that speak the same connection standard. It holds what the file cannot: which rules and decisions are in force right now, who decided each one, what would end it, and what has already been retired. Assistants write to it as proposals, and a person approves what becomes binding. It does not read your code, your repository or your data, and it does not run your agents. Keep CLAUDE.md for what the repository owns; Arroway is where the rest lives, so that people and their AIs stay on the same page.
Last updated September 24, 2026
Two different things, often described as one
Both end up in front of the model, which is why they get compared. But a context file is an artefact of a codebase: it lives in version control, travels with every clone, and describes how that code is built and which conventions it follows. Arroway is organised by project, not by repository. A project can be a codebase, a client account or a whole company, and it is read the same way from any connected tool, on any machine, by an assistant that has never seen the repository — or by one working where there is no repository at all.
What Arroway is not
It is not a tool that inspects your repository, warehouse or infrastructure and reconciles your rules against what it finds there. It does not scan code, does not run or schedule agents, and does not watch systems. It holds what people and their assistants write into it — decisions, rules, facts, handoffs, a dated log — and serves the part that is current to whichever assistant asks, ordered by the task that assistant declares. The narrowness is deliberate: a memory that also tried to be the monitor would have to decide on its own what is true, and deciding is the part kept for a person.
Where the file stops
The file is right for what a repository owns and has the same answer in every clone: how to build and run it, the layout, the conventions of the code. It stops at three things. A line in it cannot say it has been overtaken, so a superseded rule reads exactly like a live one. It cannot tell a decision a person approved from something an assistant inferred and wrote down. And it reaches only the tool that loads it, in the repository it sits in — the person deciding pricing or client terms never opens it. Those three are what Arroway is for.
What it looks like in practice
A four-person team keeps one CLAUDE.md per service, four in total. The rule that customer data never goes into a test fixture is written in three of them; the fourth service was created after the rule, and nobody copied it across. In a planning meeting the team loosens the rule for anonymised data. One file gets updated. Two keep the old wording. The fourth still has nothing. With Arroway the rule is written once, in the project, with the name of whoever decided it. When it changes, the new version points at the one it replaces and the old one leaves the read. Every assistant — in all four services, in the chat window the product manager uses, on a laptop that has never cloned any of them — opens its next session with the current version. The four CLAUDE.md files go back to describing how each service is built.
Questions people ask about this
- Do I have to stop using CLAUDE.md?
- No, and nothing in Arroway asks you to. The file stays the right place for what the repository owns. What tends to move out is the part that changes without the code changing and has a person behind it — decisions, team rules, what a client agreed to. Arroway's own repository keeps a context file for exactly the repository part.
- Does Arroway read my repository or check my code against the rules?
- No. It never connects to your repository, database or infrastructure, and it does not compare what is written against what is deployed. It only knows what people and assistants record in it. Checking a codebase is something your assistant does with the tools it already has; Arroway gives it the rules to check against, with who decided them and whether they are still in force.
- Can I see which parts of my CLAUDE.md would belong in memory instead?
- Yes. The CLAUDE.md audit on this site sorts a pasted file into decisions, rules, facts, log and noise, and estimates how much of it is the kind of knowledge a file cannot keep current. It runs entirely in your browser — the file is not sent anywhere — and the numbers it shows are an estimate from text markers, which the page says plainly.
Where this is verifiable
Product documentation on this site (How it works, Install, the CLAUDE.md audit), the answer on versioned files versus curated memory for the general comparison, and the project, sanction, supersession and read-ordering rules in the sanctioned product spec. CLAUDE.md is described here only as what it is — a file kept in a repository and loaded by the assistant working there; this page makes no claim about the built-in memory of any AI tool. Everything described here is behaviour the tools apply today, not roadmap.