Answers
Shared 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.
Last updated September 3, 2026
The thesis is decided once and retyped every week
Stage, check size, ownership target, the sectors the fund does not touch, the two exceptions the partners agreed to make — all of it was settled in a partners' meeting, and none of it lives anywhere an assistant can reach. So each partner rebuilds it in a prompt, from memory, in their own words. The version that gets applied is whichever one that person happened to type, and the parts that get left out are the ones that were hardest to agree on.
What you do once: write the criteria down and approve them
A criterion is a few lines of ordinary text — what the fund looks for, what it declines, and the reasoning without which someone would apply it wrongly. A partner writes it, or an assistant proposes one after a deal and a partner edits and approves. Nothing an AI wrote on its own counts as a criterion until a person says so. Each partner then connects Arroway to the assistant they already use, by pasting one address into its settings; from that point their AI reads the approved criteria at the start of a screen, ordered by the deal in front of it.
A thesis has a vintage, so each entry says what ends it
Funds move stage, close a vehicle, revise a sector view after a bad outcome. So each entry is stored with the condition that ends it — a date, a named trigger, an explicit revocation. When an investment period closes or the partners change a view, the entry goes back to whoever owns it instead of quietly continuing to screen deals. A superseded criterion stops being served to anyone's AI, and the old wording stays on the record for the day someone asks why the fund passed on something two years ago.
What it looks like in practice
A junior partner runs a first screen on an inbound seed deal: two founders, pre-revenue, in a sector the fund invested in twice and then decided to stop backing at seed. The assistant has no way of knowing the second half of that sentence, so it writes an enthusiastic memo — and the partners' meeting spends twenty minutes arriving at a decision the fund already made. With the criteria written down, the same screen opens holding them: this sector from Series A only, decided in November after two seed write-offs, to be revisited if a founder comes from one of three named companies. The memo now argues about whether this founder is that exception, which is the only part that needed a partner in the first place.
Questions people ask about this
- Does this mean the AI decides what the fund invests in?
- No. An assistant can propose a criterion — after a deal, it writes down what it saw the partners weigh — but a proposal is stored as a proposal and shown as one. It governs no screen until a partner approves it, and what a partner decided is never handed to the next assistant as the same kind of thing as what an AI worked out. The judgment on a deal stays exactly where it was.
- Our partners use different AI tools. Does everyone have to switch?
- No. The criteria live outside the assistants, and each partner connects the one they already use. A fund where one partner works in one tool and another in a different one still keeps a single set of criteria between them, and the partner who changes tools next year does not carry the fund's thesis away with them.
- Deal information is sensitive. What actually goes in?
- Only what you write and approve. Entries live per project, so the fund's criteria and the working notes of a portfolio company are separate collections that never see each other, and nothing arrives here from your inbox, your data room or your CRM — Arroway reads nothing you have not put in it. The practical line: the fund's criteria belong here, and a founder's confidential numbers do not need to be here for those criteria to work.
Where this is verifiable
Product documentation on this site (How it works, Install), the sanctioned product spec in the repository, and the Privacy Policy for what is stored and who can reach it. Everything described here is behaviour the tools apply today, not roadmap. This page makes no claim about the built-in memory of any AI product, and it is not investment advice.
https://www.arroway.app/en/answers/shared-memory-for-vc-funds