ZAMBO.

Direct answer

New tool, blank memory. Your working setup should be a record you carry, not a lesson you repeat.

You moved to a new AI tool, for price, for quality, because a client uses it, and the new tool knows nothing: not your clients and their quirks, not your rates and terms, not how you structure work, check it, and deliver it. So you start the familiar ritual of teaching it, in chat, over days, knowing the teaching itself now lives inside this tool and will be lost at the next switch. The mistake is treating your working setup as conversation. It is not; it is a small, stable body of facts and rules that can be written once, kept current in one place, and read by any AI at the start of work. Here is what that setup actually consists of, how to carry this move with a written record instead of a re-teaching marathon, and how to keep clients, rates, and house rules in one carried record so the tool becomes replaceable and your practice is what persists.

See the instrument first: the live demo runs one real call and hands you its verifiable receipt, no signup.

Updated November 6, 2026. First published November 6, 2026. The two delivery records quoted below are real and public, with their client decision states read back from the live service on November 6, 2026.

Step 1. Name what your working setup actually is, in four layers

Your setup feels like one blur of knowing how you work, but it separates cleanly into four layers, and only one of them belongs in a chat at all. Client facts: who each client is, their business in a line, their contacts and who decides, their preferences and sensitivities, and what is currently live with them. Commercial facts: your rates, your terms, how you quote, what you never discount, and your standard positions on deposits, revisions, and payment timing. House rules: how you work, your delivery shape, your checking habits, your writing voice where it matters, the things you always do and never do. And live project state: where each current job stands, which belongs to the project's own record, not to the setup. Write the first three layers down as the setup record. The fourth layer travels separately, job by job, and confusing the two is why setups feel too big to carry: nobody can re-teach twelve live projects, but three layers of standing facts fit on a few pages.

Step 2. Write the setup record once, in plain statements an AI can use

Draft the record as short factual statements, grouped by layer and by client, written the way you would brief a sharp new assistant on their first morning. One client per block: name, business, decider, live work in a line, and the two or three things that matter about working with them. Commercial facts as numbers and rules, not history: the rate, the deposit rule, the revision rule, stated as they stand today. House rules as instructions in the imperative: deliver in this shape, check sources this way, never promise that. Keep it honest and current rather than complete and stale; a short record you trust beats an exhaustive one from last year. Date the record, keep the master copy outside any AI tool, in your own files, and treat every AI's copy as a reading of the master, never the master itself. That one decision, the master lives with you, is what makes every future switch a copy operation instead of an archaeology dig.

Step 3. Read the two states standing facts can sit in

Standing facts only help if the work they inform lands on records that persist the same way. The record format is AER-1, published as an IETF Internet-Draft (draft-zambo-aer1-26), and each step carries a verifiable receipt: a receipt proves the saved result was not changed since it was saved. The two public records below show the decision states a job record carries, whatever tool the work ran in:

Accepted: the decision is on the record

Pending: checked, and still waiting on a decision

The accepted record, 4 steps, 4 checked, decided_by client at 2026-10-16T14:31:51.146Z, release_hint ok_to_release, shows a fact your setup record should hold in one line for that client: this job, accepted, on that date, on those terms. The pending record, 5 steps, 5 checked, release_hint hold, is the other line every practice needs at a glance: work finished, decision outstanding, do not start the next stage. A new AI briefed with setup facts at this grain, per client, per job state, can act sensibly in its first hour: it knows what is settled, what is waiting, and what your rules say to do about each. Briefed with a chat export instead, it knows a lot of wording and none of the states, and it will cheerfully offer to chase a job that was accepted a month ago or re-quote a rate you changed last season.

Step 4. Bring the new tool up in one session, from the record

Run the move as a single deliberate session rather than a week of drips. Open the new tool, give it the setup record as its first material, and have it work from the record in front of you: summarise each client block back to you, restate your commercial rules, and flag anything ambiguous or contradictory. Correct the record, not just the chat, wherever its reading exposes a gap, so the master improves with every switch. Then run one small real job end to end in the new tool, a minor client task with a known right answer, and check how it applies your rates, your delivery shape, and your checking rules in practice. Gaps found on a small job are tuition; gaps found on a live delivery are damage. By the end of that session the new tool is not an apprentice you are raising; it is a reader working from your record, which is the relationship you want with every tool from now on.

Step 5. Keep the record current at the moments facts change

A carried setup dies by staleness, not by switching. Facts change at identifiable moments, so attach the update to the moment: when a rate changes, the master record changes the same day, before the next quote goes out in any tool. When a client relationship changes, decider, sensitivity, live work, update their block when you close the job that changed it. When you adopt or drop a house rule, write it in the week you decide it, while the reason is still fresh enough to state in a line. A monthly ten minute pass over the record, deleting what is dead and dating what changed, keeps the whole system trustworthy. What makes this cheap is that the record lives in one place; updating five tools' memories was never going to happen, which is precisely why tool held setups rot. One master, many readers, updates at the moment of change: that is the whole maintenance regime.

Step 6. Keep live project state on job records, and let the setup stay small

Resist loading live project detail into the setup record as the practice grows; that way lies the three hundred line brief nobody maintains. The setup holds standing facts only: clients, rates, rules. Each live job holds its own state on its own wrapped record, steps, decisions, open items, which any AI picks up when it works that job. A new tool then comes up to speed in two readings: the setup record for how your practice works, and the job record for the specific work in front of it. The same split serves clients directly. When a client asks what you hold on them, the answer is one block of standing facts plus their job records, findable and showable, rather than a confession that it is somewhere in a chat history in a tool you no longer use. Small setup, per job records, master in your hands: switch tools whenever a better one appears, and lose nothing but the login.

What a record does not settle

A verifiable receipt proves the saved result was not changed since it was saved; it does not move a platform's private chat history into a new tool, teach a new AI your judgement, or keep your setup record current for you. The four layer split, the dated master record in your own files, and updates at the moment facts change are what this answer stands on. This page is general information, not legal advice.

Frequently asked questions

I switched to a new AI tool and it knows none of my clients, my rates, or how I work. How do I carry my working setup over instead of re-teaching everything?

Write the setup once as a record you own, in three layers of standing facts, and have every tool read it. Client facts, one block per client: business in a line, who decides, what is live, and the few things that matter about working with them. Commercial facts as they stand today: rates, deposit and revision rules, payment terms, stated as numbers and rules. House rules as plain instructions: delivery shape, checking habits, what you always and never do. Keep the dated master in your own files, outside any tool, bring the new tool up in one session by having it read the record, summarise it back, and run one small real job against it, and update the master at the moment any fact changes. Live project state stays on each job's own record, so the setup stays small enough to maintain and the tool becomes a replaceable reader of it.

Can I export my memory or chat history from the old tool and import it into the new one?

Sometimes, partly, and it is worth doing when the route exists, with clear eyes about what crosses over. Exports move material: past conversations, saved outputs, sometimes a list of remembered facts. They do not move authority: the new tool receives text, and it cannot tell which statements in that text are your current rates, which client facts were corrected later, and which decisions were reversed, because in a chat all of those wear the same clothes. Use any export as source material for building the setup record in this page, checking each standing fact as you lift it, rather than as the setup itself. The record you write and date is the first version of your setup that is actually portable, because it was written to be read by a stranger, human or AI, and that is what a new tool is.

What should I leave out of the setup record?

Leave out anything that is not a standing fact, and anything you would not want read aloud in the wrong room. Live project detail belongs on job records, not in the setup: it bloats the record, goes stale fastest, and is the main reason setup documents get abandoned. Personal data about clients beyond what the work needs, private contact details, credentials, and anything given to you in confidence should stay out or be held in your own separate notes with the setup record pointing at them rather than copying them. Also leave out history: the record states your rates and rules as they stand today, dated, and a short change line at the foot is enough archaeology. A setup record is a briefing, not a memoir. If a line would not change what a new AI does in its first week, it does not earn its place.

How do I handle a fact that differs per client, like rates or terms?

State the default once, then state the exception inside the client's block, and make the exception carry its reason and date. Most practices have one real rate card and a handful of client specific arrangements: a legacy rate for a founding client, different payment terms for a large company, a standing discount that was agreed for a defined scope. Written as default plus named exceptions, a new AI quotes correctly in both cases and can explain the exception if the client queries it. Written as a single blended description, it will eventually quote the legacy rate to a new client or the standard rate to the founding one, and both errors look like carelessness. The per client block is the natural home for exceptions precisely because the AI reads the block of the client it is working for; keep the default prominent so a client with no exception gets the standard terms without hesitation.

How do I carry my setup so the next switch costs an hour, not a week?

Hold the standing facts where any AI can read them on pickup, and keep the job state on records that travel the same way. When your work runs through Zambo, facts you save with the memory tools sit behind one endpoint that any of your AIs can remember into and recall from under the same connection, each job wraps as a handover with a verifiable receipt on every step under AER-1, published as IETF Internet-Draft draft-zambo-aer1-26, and picking work up in another AI is always free. New users get 20 calls and 1 wrap-up free every day, and picking up work in another AI is always free. For $0.99, run the new tool's first real job that way and let it start from your record instead of your recollection: get the Day Pass. A Day Pass is unlimited for 24 hours, no subscription. Ongoing use is the Zambo Pass at $29 a month or $290 a year, and $19 a month for $ZAMBO holders, which adds private handover links, your branding, and the approval record as PDF and JSON files.

Related