Answers about your own account

Most people never find out this exists, so here it is up front: signed in — on every plan, Free included — Onflow AI does not only answer from the guide. It can answer about your own account — the reference you just pasted, the reliability event that cost you points, whether that top-up is still refundable, which channels are worth approaching.

It is read-only, it is only ever your account, and it is governed by four rules that do not bend. This page is those rules and everything the account lens can read. Since September 2026 a second layer sits on top of it — reads and actions across Boost, Paid Promotions, the Subscriber Exchange and Cross Promotion, each proposed first and carried out only after you confirm — and that one has its own page.

The short version
Where	The Onflow AI chat in the help widget, on any page.
Who	Every signed-in member, on every plan. Signed out, the chat helps with the guide and the policies only — it never reads an account, and there is no in-chat verification. Acting on your account (cancelling, confirming, deleting, the data report) is a separate capability on Plus and above — see What the assistant can do for you.
What it costs	Nothing extra. An account-aware turn spends the same message allowance as any other turn.
What you do	Just ask. Paste the reference, or ask about your score, your refunds or partner suggestions.

What it can look up

Ask about	What you get
A reference you paste	What that OF… id names, when it was created, and its own key details — the amount and status of a top-up, the size and direction of a wallet movement, the status of an order, the stored subscriber count of a channel. Up to three references in one message.
Your reliability score	The current score out of 10 and your recent ledger — every change with its date, its size, and a plain-English cause, plus whether it was reversed, is under appeal, or is still appealable.
Your refunds	Your recent top-ups checked one by one against the live Refund Policy, each marked eligible or with the reason it is not — and, when one genuinely qualifies, confirmation of the refund the policy provides.
Partner suggestions	Real, opted-in cross-promotion channels close to your own channel in size and reach, two to four of them, with reasons.
"What can you do for my account?"	A straight answer, drawn from a fixed description of these capabilities rather than from the model's imagination.

The four rules

These are not guidelines. They are enforced in the code, in the database, and in the wording of the replies.

Your identity is proved, never claimed

Nothing you say about who you are counts. The layer speaks for exactly one account: the one on your signed-in session. There is no other way in — a conversation that is not signed in gets no account data, whatever is typed into it, and a conversation that was started under a different account gets none either.

Your data only — and a stranger's reference reads the same as a made-up one

A reference is described only when the registry says your account owns it. Everything else — a reference belonging to somebody else, and a reference that never existed — produces the identical reply:

I don't have anything I can share about OFB-204-1188. I can only discuss references that belong to the account you're signed in with — double-check the reference, or ask our team below if you believe this is a mistake.

That sameness is deliberate. If a foreign reference produced a different answer from a nonsense one, the chat would become a way of testing whether an id exists and who holds it. It never confirms and never denies.

Signed out, no lookup happens at all — you get a pointer to sign in, and nothing more.

Every lookup is read-only, and the database enforces it

Every one of these lookups runs inside a read-only database transaction. Not "the code is careful not to write" — the database itself refuses a write from inside that transaction, so a bug, or text engineered to look like an instruction, still cannot change anything.

It explains; it never decides

Account data reaches the model as a typed block of facts it may only restate. Anything with a consequence attached — the refund confirmation, the "your plan does not include this" notice, the no-information line above — is fixed text written by the server, not by the model.

Which means, concretely:

• It cannot lodge a refund, approve one, or move a single cent.
• It cannot change a plan, edit your account or unlock anything.
• It cannot overturn a reliability penalty; it can only explain it — and, on your typed confirmation, file the appeal for you.
• Routine tasks — cancelling an order that has not started, accepting a booking, answering a partner request — are a separate layer that acts only after you type CONFIRM in the chat. See What the assistant can do for you.

The one promise it can make, and its limits

When a top-up currently qualifies for a refund, Onflow AI will say so and flag the conversation for the team — because eligibility was checked live, against the same rule the refund system itself uses. It still cannot lodge the request: that step is yours, from Wallet → Top-ups. Eligibility is re-checked when you lodge it and again at review, so a promise in the chat is a live reading, not a decision already made.

Who gets what

Visitor	What the chat does
Signed out	Answers from the guide and the policies. No account data, no lookups, no in-chat verification — a question about "my account" is answered with a pointer to sign in.
Signed in — any plan, Free included	Everything on this page: your identity card, your references, your reliability ledger, your refund standing, partner suggestions, and the read-only cards under What the assistant can do for you.
Plus and above	All of that, and the chat can act: cancel or confirm things for you, start or stop an account deletion, send your data report, and confirm a qualifying refund to the team. Below Plus the chat names the plan and tells you where to do it yourself — and for a deletion or your data report, which have no button on the site, that is [email protected] from your account's address, or the contact form.

The team can move the rungs or switch either layer off from the admin console; the chat always tells you which plan a thing is on when yours is below it.

References it recognises

Paste any Onflow Ads ID, on its own or mid-sentence, in any of the usual shapes — OFD-204-7831, OFD2047831, upper-case with spaces. Older-format references from saved emails still resolve, and it will tell you what the modern reference is.

Every owned reference gets its name, its creation date, and whether it has been retired. Many carry more:

Reference	What it adds
OFD-… deposit, OFI-… invoice	The amount, the provider, the status, and any refund status
OFR-… refund request	The amount, the status, and when it was decided
OFW-… withdrawal	The amount and the status
OFT-… wallet transaction, OFBT-… boost-credit transaction	Credit or debit, the amount, and what kind of movement it was
OFRE-… reliability event	The exact change, the plain-English cause, and the score it left you on
OFCH-… channel	The title, the @username and the subscriber count held on the channel record
OFB-… boost order, OFO-… marketplace order	The status
OFC-… cross-promo campaign	The campaign name and the status
OFSE-… subscriber-exchange campaign	The status
OFS-… support ticket	The subject and the status

Everything else in the id catalogue still resolves — you are told what it is and when it was made.

The channel figure it reads back is the stored one, and it is not labelled

For an OFCH-… channel the chat reads the subscriber count stored on the channel record and states it flatly. That stored number is usually a reading taken from Telegram — but it is also the owner's to edit, so it can be a figure someone typed rather than one anybody measured. The chat draws no distinction; the channel's own dashboard does, labelling any figure it could not measure live "As stated by the owner — not measured by us". Where the two surfaces seem to disagree about how solid a number is, the dashboard is the one telling you where it came from.

Your reliability score

Ask why your score changed and you get the ledger, not a shrug:

• The current score, out of 10.
• Your recent changes, newest first — each with its date, the exact amount it moved, and what caused it in plain language: "a paid post was not published inside its agreed window", "you delivered a paid promotion cleanly, on time and in full", "a clean streak earned back some score".
• The score after each change, so you can follow the arithmetic.
• Its standing: since reversed, appeal under review, or still appealable from the reliability page.

If the history genuinely cannot be read at that moment, it says so and points you at the reliability page — it will not present a failed read as an empty ledger.

Background: Your reliability score and Appealing a penalty.

Your refunds

Ask about a refund and it checks your recent top-ups against the live Refund Policy — the same rule the refund system itself applies, read at that moment.

You get the policy in one paragraph — a top-up is refundable only within 12 hours of crediting, only while fully unspent, always to an external destination and never back into the wallet, with a flat 18% platform fee plus gateway fees deducted; review takes up to 5 business days and an approved payout a further 7–14 business days — and then each recent top-up marked either eligible for an external refund right now or with the reason it is not.

When exactly one of them qualifies, or you named the one you meant, it adds the confirmation:

Good news on OFD-…: it currently qualifies for a refund under our policy, and I've flagged this conversation for our team so they expect your request. To start it, open Wallet → Top-ups and choose "Request refund" on that top-up…

Read the rest of that message carefully — it carries the fees, the two timelines, and the one thing that will cost you the refund if you get it wrong:

Do not spend from your wallet before you lodge it

Eligibility is re-checked when you lodge the request and again at review. Any spend from the wallet ends that top-up's eligibility, and the chat saying "eligible" a minute ago does not preserve it. If you want the refund, lodge it first and spend nothing in the meantime.

Two more things worth knowing: a crypto top-up will ask you for a fresh USDT BEP-20 address to receive the refund, and the confirmation is made once per top-up per conversation — the standing stays visible afterwards, but the team is not pinged twice for the same thing. The full walkthrough is Requesting a top-up refund.

Partner suggestions

Ask about cross-promotion partners and it works from real listings, never from invention:

• It takes your largest connected channel as the reference point.
• It reads the partner directory — active, opted-in listings belonging to other people.
• It ranks them by closeness to you in subscribers and average views, and suggests two to four, each with a reason: their size, their typical reach, their theme and their own blurb.
• You send the requests yourself, from Find Partners.

If no channel is connected to your account yet it says so and tells you to connect one first. If the directory has no other listings right now, it says that too rather than making some up.

What it will never do

• See anyone else's account. On any plan. Ever.
• Write anything on its own. No money moved, no plan changed, no penalty lifted. The only writes it ever makes are the assisted actions you confirm word for word.
• Promise beyond the published policy. The single confirmation it can give restates the policy exactly and alerts the team.
• Answer an account question for an unproved visitor. It offers the two ways to identify yourself and stops there.

If something goes wrong

What you see	What it means	What to do
"References like OFB-… are private to the account that owns them…"	You are not signed in	Sign in and ask again
"I can answer questions about your own account once you're signed in…"	You asked something account-shaped while signed out	Sign in
"Doing things inside your account from this chat … is part of the Plus plan and above."	You asked the chat to act, on a plan below Plus	Do it yourself on the page it names, or upgrade at /pricing
"I don't have anything I can share about OF…"	That reference is not yours, or is not a reference	Re-read it from the email or page you copied it from; ask a person if you are sure it is yours
An account question answered without a specific ledger, refund check or partner list	Your account summary rides every message, but those deeper lookups run only when you name the topic or paste a reference	Paste the exact reference, or name the topic — your reliability score, your refunds, partner suggestions

Privacy

On the plans that include account-aware answers, every message you send carries a typed summary of your own account — assembled by the server and capped in size — to the AI model provider, so the answer is written from your real record rather than a guess. It has three parts: an account summary with your identifiers masked (your email is shown only partly and your Telegram user id never), a count of the entries in each part of your record, and a few of the newest entries from the parts your question is about. The deeper lookups — the reference you pasted, the events behind a score change, the refund window of your top-ups, the channels being suggested — are added when you ask for them.

It is your data and only ever yours: no other member's account identifiers, contact details or messages are read — your own orders show the channel you booked, and partner suggestions come only from listings whose owners opted into the public directory. The summary covers your display name, Onflow Ads ID, plan, balances and lifetime top-ups, referral code and count, reliability score, when you joined and how you signed up, and any freeze or outstanding balance with its dates; when a question names nothing in particular, your most recent activity rides along instead. The read is read-only at the database itself, and the model may only restate what it is given. In this feature, account passwords and other sign-in credentials, two-factor secrets and codes, IP addresses and devices, identity documents, and our internal notes and risk signals are never sent; your payout destination and the payment instrument you paid with are never sent to a model provider by any feature. The assistant also declines to share sensitive details even when asked. The Privacy Policy sets it out in section 12.2, and the per-account AI switch described in Chatting with Onflow AI turns all of it off.

Related

• Chatting with Onflow AI — the conversation, the allowance and the handover to a person.
• What Onflow AI knows — the published documents behind every other answer.
• Your Onflow Ads IDs — what each prefix names, and why an id is safe to share.
• Requesting a top-up refund — the step only you can take.
• Your reliability score — the ledger the chat reads out.