One decision, four moments
How Arroway keeps a decision alive
Follow one decision as an AI reads it before working, records what changed, and leaves people able to correct the result. That is the Arroway loop.
It reads before it acts.
At the start of a task it pulls what the team has decided about that subject: the rules that always apply, the decisions closest to what you're doing, and what your teammates have been working on lately.
It writes when it finishes.
Not the conversation — the residue. What was decided, what was found, what changed, what it cost. The thing a person arriving next week would need in order not to redo the work.
One decision, four moments
One decision, four moments
- 01
Read what is already decided.
Before a task begins, the AI receives the rules and decisions that apply to this work.
Details
It starts with the team’s context, not a blank guess.
- 02
Work with that context.
The AI prepares the next step using the decision instead of reopening it from scratch.
- 03
Remember what changed.
When work leaves a durable result, Arroway records it for the people and AIs who come next.
Details
A new Arroway begins empty on purpose: start from the documents, rules and decisions that already govern the work.
- 04
Correct in the open.
A person can correct or reject a proposal at any time. Review is a quick veto, never a queue that blocks the work.
A public demo project
See a real, curated memory before you sign in
This is a safe demonstration project. These are the same kinds of rules, decisions, facts and proposals a team sees while it works.
The fictional team writes in English, so its records stay exactly as they were written. This public view never exposes a customer project, a person or a source link.
Demo project
Atlas — checkout rebuild
Money is an integer in the currency's minor unit
An assistant drafts money movement; a person sends it
Every write to the payment provider carries an idempotency key
Subscription state lives in the provider, not in our tables
What shows up for you
What shows up for you
It is a veto, not a gate. Arroway keeps working whether you look or not; review is where you correct a proposal before it quietly becomes a bad habit.
In the last three negotiations, anything over 15% was approved by Rafa before it reached the client. It looks like a rule, but nobody ever wrote it down.
Decided in January: below 10 seats the discount eats the margin and does not shorten the cycle.
You don't have to wait for the weekly pass. Everything is editable at any time, and you can write a rule yourself without waiting for an AI to propose it — useful on day one, when the team's rules exist but no AI has been around long enough to notice them.
The first use starts empty — on purpose
The first use starts empty — on purpose
If the work already has memory, ask the AI to read its Arroway before acting. If it is empty, begin with the documents, rules and decisions that already govern the work. The AI proposes what to record; you can correct what does not belong.
It reads before it acts.
At the start of a task it pulls what the team has decided about that subject: the rules that always apply, the decisions closest to what you're doing, and what your teammates have been working on lately.
It writes when it finishes.
Not the conversation — the residue. What was decided, what was found, what changed, what it cost. The thing a person arriving next week would need in order not to redo the work.
It labels what is a proposal.
A human decision and an AI inference do not look the same. Inferences stay visible as proposals until a person changes or rejects them.
It notices when two things collide.
If something it is about to write contradicts something already there, it says so instead of quietly stacking a second version on top.
What Arroway never touches
Your AI sends what it writes. It does not send your conversation.
The full detail is in the privacy policy, which is written the same way this page is: what actually happens, including the parts that are inconvenient.
Questions people actually ask
Arroway orchestrates people and AI agents around the decisions in force.