Answers
Answers about shared memory for people and AI
Start with the question in front of your team. Each answer explains one part of how shared context reaches the AI before work starts.
Explore the answers
Shared memory for teams that work with AI
Shared memory for a team is one place — outside every assistant — where the team's decisions, rules and daily record are written down, and which each person's AI reads before it starts a task and writes to when it finishes. Arroway is that place. The memory belongs to the project, not to anyone's chat window, so the second person's assistant starts from what the first person already settled, instead of from a blank.
Read the answerShared context across Claude, ChatGPT and Codex
One connection, and every assistant reads the same project. Arroway runs as a server your AI tools connect to — Claude, ChatGPT, Codex and any other tool that speaks the same connection standard — so they all read and write one memory. What you settled in one tool is already there when you open the next, and changing tools does not start you over.
Read the answerAI memory a person approves before it counts
Human-sanctioned memory means a person decides what becomes a rule, and the AI cannot promote its own guess to that status. In Arroway an assistant writes freely as it works, but anything it worked out on its own is filed as a proposal. A person turns proposals into standing rules — or edits them, or throws them out — on a review screen, after the fact. The work never waits on that review.
Read the answerHow team decisions reach the AIs doing the work
A decision reaches an AI when it is put in front of that AI before it acts — in the read the assistant makes at the start of a task, not in a document someone hopes it will find. Arroway keeps decisions as material of a project, marks the ones that must apply always, orders the rest by the task in front of the assistant, and says in the same breath how much fit and what did not.
Read the answerHow to install and use Arroway in the main AI ecosystems
You pick the tool you already work in, connect Arroway to it, and start working. The installation page asks one question — where do you use AI? — and shows one route, not a menu of every possibility. It does not ask you to sign in until the moment a sign-in is actually needed.
Read the answerShould team knowledge for AI live in versioned files, or in curated memory?
Files in the repository — a CLAUDE.md, a folder of skills, notes committed next to the code — are where most teams put what their AI should know, and they are right for what a repository owns: how to build it, the conventions, the shape of the code. What a file cannot carry is the part that changes and has a person behind it — which rule is in force today, who decided it, whether it is a settled decision or an assistant's guess, and whether the model actually received it. Arroway holds that part: written by people and their assistants, approved by a person, ordered by the task, and delivered with a count of what fit.
Read the answerWhere does team memory fit in an AI-native software lifecycle?
The AI-native loop is usually described as a chain of artefacts: an intent written down, a plan derived from it, code produced against the plan, and a review closing the circle. That describes how one piece of work travels. It does not carry what stays true between pieces of work — which rule is in force, who decided it, why an approach was abandoned, and what the people who never open the repository already settled. That standing layer is what Arroway holds, and it is put in front of the assistant before it plans anything.
Read the answerYour CLAUDE.md outgrew a page. What comes out of it?
The advice to keep a CLAUDE.md under a page is good, and it is hard to follow because two different kinds of knowledge end up in the same file. One belongs there for good: the commands, the conventions, the mistakes an assistant keeps repeating in this codebase. The other is what a person decided — which rule is in force, why an approach was dropped, what may be promised to a customer — and it grows without limit, because decisions never stop being made. The cut is not between important and unimportant. It is between what the repository owns and what the team owns.
Read the answerShared memory for law firms: the house position, in every partner's AI
A firm's position on a recurring question belongs to the firm, but each partner's AI only knows what that partner typed into it. Arroway is a shared place, outside every assistant, where those positions are written down and approved by a person at the firm. From then on, when any partner's AI starts work on a matter, it reads the positions that apply before it drafts anything. Nobody re-explains the house view in a prompt, and nobody finds out that a colleague answered the opposite after the advice has already gone to the client.
Read the answerAI memory and client confidentiality: how one client's matter stays out of another's
Memory lives inside a project, and a read never crosses from one project to another. A matter, a client, a practice group — each is its own project, and what is written there is served only to the people who belong to it, on connections that identify them personally. Nothing is shared by default, nothing spreads because two matters look alike, and what should never be within an AI's reach is kept out by the only control that actually holds: a person decides what gets written down at all.
Read the answerThe partner on the account changes. What does the next one inherit?
What a successor should inherit is the decisions the account already carries — not a folder of e-mails to mine and a calendar of calls nobody wrote up. With Arroway the account remembers instead of the person: each decision taken with that client is written down as it happens, approved by a person, and read by the next partner's AI the moment they open the work. The handoff meeting stops being the only transfer — and stops being the place where whatever nobody thought to mention is lost.
Read the answerShared memory for VC funds: the thesis your AI already knows
A fund's thesis and screening criteria are settled once by the partners and then applied to every deal that arrives — but the AI doing the applying only knows what somebody pasted into it that morning. Arroway is a shared place, outside every assistant, where the thesis, the criteria and the hard lines are written down and approved by a partner. From then on, when any partner's AI opens a deal, it reads the criteria that apply before it forms a view. Nobody retypes the thesis into a prompt, and nobody finds out afterwards that a screen ran against last year's version of it.
Read the answerOne rule, four prompts: why your AI automations drift apart
When the same rule is pasted into four automation prompts, you do not have one rule — you have four copies that will stop agreeing. Correcting it means finding every file that carries it, and the copy you miss keeps running. With Arroway the rule is written and approved once, in a shared place outside the prompts, and every routine reads it when it starts. The correction is a single edit, and no prompt is left holding a version nobody remembered to update.
Read the answerHanding work between your own AI routines, without re-narration
Scheduled AI routines usually pass work to each other by retelling it — a note in a file, a message in a channel, last night's transcript — and a retold fact drifts. With Arroway a run that stops unfinished leaves an explicit handoff: where it stopped, the one next step, the risk left open, and who it is for. That handoff is served first to whatever reads next, and it stays there until someone closes it. Nothing is inherited as a summary of a summary, and nothing quietly falls off the list.
Read the answerI 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.
Read the answerWe are switching AI tools. Do we lose what the team already decided?
Not if the decisions were never inside the tool. What a team loses in a migration is whatever it kept in a vendor's own memory, its settings and its per-tool configuration files — those belong to that tool, and no export makes them arrive intact somewhere else. Arroway holds decisions in a different place: a server your tools connect to. The new assistant reads the same sanctioned set the old one read, on its first day, so the switch costs you configuration instead of context.
Read the answerHow does an AI agency keep one client's rules out of another client's work?
By keeping each client in its own project, and the agency's own way of working in a separate one. Memory in Arroway is scoped to a project and never global: an assistant working on an account reads that client's sanctioned rules plus the agency's house method, and nothing from any other client — because the other accounts are not in scope for that read, not because someone remembered to leave them out. The same person and the same assistant can work six accounts in a day without the sixth inheriting the voice of the first.
Read the answerNew associates using AI: how the partner's call reaches them first
An associate working with AI answers with whatever the AI knows, and what it knows is what that associate told it — not what the partners settled on the same question two years ago, in a matter the associate never touched. Curated memory changes the order: the position the firm already sanctioned is served to the associate's AI when the work opens, before a first draft exists, and whatever that assistant works out on its own is stored as a proposal rather than as the firm's answer. The associate starts out knowing where the house stands, instead of someone catching the contradiction in review.
Read the answerShared memory for multi-partner consultancies: what the firm standardises, what stays yours
In a consultancy each partner has their own clients and their own way of running an engagement, and both are worth keeping — so the useful question is not how to make every partner's AI answer the same way, it is which part of the answer belongs to the firm. Curated memory draws that line explicitly: what the firm sells — the method, the standards, the things it will not do, the register it writes in — is written once and sanctioned, and it reaches any AI any partner connects. How each partner runs their own account stays in their own project, unstandardised on purpose.
Read the answerWe tried a shared doc of client rules and nobody updated it
The shared document did not fail for lack of discipline; it failed because keeping it current and reading it were both extra gestures, and an extra gesture loses to billable work every time. Curated memory removes both. The assistants already inside the work write what was decided at the moment it is decided, and the norm is read when an assistant opens the work rather than when somebody remembers to look it up. What is left for a person is a short review — approve or correct what was written — which is a few minutes over work already done, not a chore waiting to be done.
Read the answerWhy not just run RAG over our repository instead of another AI memory tool?
Because retrieval and sanction answer different questions. RAG finds the passages most similar to what was asked, out of everything it indexed — including the decision the team reversed last quarter, the design that was never built, and the two documents that contradict each other. It has no way to say which of them is in force, because nothing in an index carries that. Arroway holds a much smaller set: the rules and decisions a person on the team approved, each with the name of who approved it and the condition that ends it. An assistant reads that set before it acts. Retrieval still finds text for you; it is not what tells the assistant which text still counts.
Read the answerHow does an AI agency change the person on a client without losing what the account knows?
By keeping what the account knows in the account's own project, written as it is decided, instead of in the head of whoever is currently on it. When someone is taken off the client they lose access to that project, and everything they recorded there stays — under their name, with the date and the reason. The person taking over reads the client's rules in force on their first day, and their assistant reads the same set. The only thing that leaves with the person is what nobody ever wrote down, which is why the writing happens during the work rather than in a handover the week someone goes.
Read the answerHow do we show a client that our AI followed their rules instead of inventing one?
By showing the record the assistant read, rather than a summary written afterwards. Every rule in force in a project carries what was decided, when, and which person approved it — and a rule an assistant proposed that nobody approved is visibly still a proposal. So the answer to a client asking where a rule came from is a dated line with a name on it. What the record does not do is prove that one particular deliverable came out of it: it shows what governed the work, and your own review is what ties a specific piece to it. Claiming more than that would be claiming more than the record holds.
Read the answerCan we recommend an AI memory tool to our portfolio without becoming its support desk?
Yes, because nothing about it routes through you. Each company signs up on its own, in its own account, and connects the assistants its own people already use — the fund holds no licence, no admin seat and no key on their side. The honest half of the recommendation is what installing does not do: connecting takes minutes and works unattended, but nothing changes until someone at that company writes down the first few decisions their assistants should be reading. That step belongs to them, and a recommendation that leaves it out produces a company that installed the tool and never used it. There is no referral link and no reseller arrangement to sign — the recommendation is a link and an opinion.
Read the answerEach partner runs diligence with a different AI and the notes never reconcile. What fixes that?
Putting what the diligence has already established in one place that every partner's assistant reads, instead of in two sets of notes that only meet at the committee. Each finding goes in once — the revenue figure that was confirmed and the document that confirmed it, the customer reference that came back weak, the question still open — and from then on any partner's assistant opens the work already holding it, whichever tool that partner uses. What this removes is not disagreement between partners, which is the job: it removes two assistants reporting different numbers with equal confidence because each one read a different half of the material.
Read the answerOur scheduled AI routines start from zero every run. How do they keep the thread?
By writing what each run established into a place the next run reads before it does anything. A scheduled routine has no memory of its own past runs — every trigger is a fresh context — so tonight's run re-derives what last night's already worked out, and either reaches the same conclusion at the same cost or reaches a different one with nothing there to contradict it. The habit that fixes it has two halves: the run opens by reading the rules in force and what recent runs recorded, and it closes by recording what it found in a line or two, including when it found nothing. A run that recorded nothing is indistinguishable from a run that never fired, and that is the failure worth designing against.
Read the answerI keep a second brain for myself. What is the second brain for the AIs my team uses?
It is a shared record the AIs read before they act, holding what your team has actually decided. A personal second brain is written to be re-read by its author: you wrote the note, so on opening it you silently supply everything it left out — that this one was a maybe, that it stopped being true in March, that the tone rule is for client work and not for internal drafts. An AI has nobody sitting beside it to supply any of that. So the second brain for your AIs has to say out loud three things a personal one never needs to: what is a decision and what is still somebody's proposal, when each thing stops being true, and who put it there. That is what makes it safe for a tool to read the record and act on it without checking with you first.
Read the answerWe run several AI agents and people on the same work. Is an orchestration tool what we need?
Two different problems get called orchestration, and only one of them is about scheduling. Ordering the work — what runs when, what waits for what, what retries after a failure — is what an orchestrator does, and if that is what is broken, get one. The failure most teams actually hit looks nothing like it: two agents run correctly, at the same time, on different premises about what the team decided. Nothing is late, nothing is stuck, nothing retries, and the two outputs contradict each other. That is not fixed by deciding who commands whom. It is fixed by every agent and every person reading the same decision in force before acting, and recording what they did. A queue coordinates order; a shared record coordinates agreement — and a team with several agents usually needs both.
Read the answerWon't a big enough context window fix this without an AI memory tool?
No, because the two solve different problems. A larger window raises how much text an assistant can hold at once. It does not say which of that text is still in force, who approved it, or what has since replaced it. Put a team's whole history in front of a model and you have handed it the decision that was reversed in March alongside the one that replaced it, with nothing marking which is which. Arroway holds a much smaller set — the rules and decisions a person on the team approved, each carrying the name of who approved it and the condition that ends it — and an assistant reads that set before it acts. Capacity was never the missing piece. A statement of what currently counts was.
Read the answerIsn't out-of-date AI memory worse than having no memory at all?
The risk is real and it is the right thing to ask about: an assistant that states a dead rule with confidence does more damage than one that asks. What decides it is whether anything ends a memory other than someone remembering to delete it. Here every rule and decision is written together with the condition that ends it — the revocation, the date, the event that makes it false. When that condition is met the memory is retired and stops being served, and a decision that replaces an earlier one names the one it replaced instead of sitting beside it. The failure you are describing belongs to stores that only accumulate. Expiry is part of the record, not a cleanup someone has to schedule.
Read the answerWhy does the AI drop a rule we agreed on five messages ago?
Because the rule only ever existed inside the conversation. Anything agreed mid-session sits in the same window as everything said since — the code that was pasted, the errors that came back, the tangent about something else — and the model's own defaults are what it falls back to as that competition gets crowded. A rule that has to survive being told once is not stored anywhere; it is being remembered by a process nobody asked to remember it. Here a rule is written down once, outside any conversation, and every assistant reads the current set before it starts. It does not depend on the session recalling it, which is also why it survives the next session, the other tool, and the teammate who was not there.
Read the answerWhose memory is it when it lives on the AI vendor's server?
It comes down to one test: can you take it out and read it without their software. A memory feature built into an assistant is part of that product — what it keeps, how long it keeps it, in what shape and under which policy are the vendor's decisions, and they can change without anyone asking you. Arroway is a separate record that belongs to your team. Its contents are text your people wrote, exportable in full at any time, read by whichever assistant is working rather than by one of them, and governed by rules the team writes. The question is not where the bytes physically sit — ours sit on a server too. It is who decides what the record contains, and whether it survives you changing your mind about the tool.
Read the answerIsn't AI memory without a write policy just a faster way to hallucinate?
Yes, and it is the right objection to make about most of what is sold as agent memory. If anything an assistant produces can become a stored fact, the store turns into the place where one confident mistake is preserved and handed to the next run as settled. Better retrieval makes that worse rather than better, because it makes the wrong entry easier to find. What breaks the loop is a step between writing and governing: here an assistant can propose, and only a person on the project can make a proposal binding. Until someone approves it, it sits in a review queue as a suggestion, and no assistant reads it as a rule.
Read the answerDoesn't a knowledge graph solve this by linking the facts together?
It solves a real part of it, and the part it solves is not this one. A graph is good at identity: it can work out that "the billing service", "billing-svc" and "the payments backend" are one thing, which a flat list of notes cannot. What it does not carry is authority. Link the facts about that service and the contradictory versions now arrive together — last year's retry policy beside this quarter's, connected, consistent and equally present, with no edge saying which one is in force. Arroway answers that question first: one entry is current, it names who approved it, and the one it replaced is retired with the reason attached.
Read the answerOur AI writes 400-line handoffs and the next agent reads half of one. What fixes that?
The fix is not a shorter prompt or a stricter style rule — it is separating what TRAVELS from what is kept. Length is what the reader pays, and what the next agent needs fits in one authored line. In Arroway every entry carries an essence written by hand: the operative state in a single sentence, never a copy of the body's opening. That line is what a read serves. The full body stays recoverable and comes back only when someone asks for that entry by name. So the record can be as thorough as the work deserves without the next twelve reads paying for it — and because the line is authored rather than truncated, it says the thing instead of the first 200 characters of the thing.
Read the answerThere are already dozens of AI memory tools. Why would we evaluate another one?
Because the thing they are crowded on is not the thing that is missing. Almost every tool on that list answers the same two questions — how text gets stored, and how it comes back — and on those two they are genuinely hard to tell apart, which is exactly what makes the category feel saturated. The question none of them answers by default is who decided that a stored statement is what the team stands behind, and what happens to it when someone changes their mind. That is governance, not retrieval, and a crowded retrieval market says nothing about it. The honest version of the objection is that you cannot tell these tools apart on the axis you were comparing them on — which is true, and is a reason to change the axis rather than to stop looking.
Read the answerI don't want AI tools remembering me. I keep context per project. Isn't that enough?
Keeping context per project is the right instinct, and it is the one this is built on — what gets remembered here is the project, never the person. The gap is not in the scope you chose, it is in what holds it. Per-project context that lives in a file you edit works while you are the only one reading it and the only one writing it; it starts costing the moment a second person, a second machine or a second assistant touches the same project, because nothing in the file says which version is in force or who decided. So this is not a personal-profile product with a project setting. The unit is the project's decision, a person sanctions it, and whoever opens that project next reads it — another teammate, another tool, or you in three weeks.
Read the answer