Skip to main content

Chatting with Onflow AI

Press Start new support chat and you are in a live, multi-turn conversation. Ask anything about Onflow Ads, ask follow-ups, change the subject, have it carry out a routine task for you, and — whenever you want — have the whole conversation handed to a person without repeating yourself.

What it answers from is on What Onflow AI knows. What it can say about your own account is on Answers about your own account. What it can do for you is on What the assistant can do for you. This page is the conversation itself.

Before you start​

  • The chat is open to everyone, signed in or not — signed out it answers from the guide and the policies; signed in it can also read your own account, on every plan.
  • It is metered per message — yours and the assistant's each count one. Paid plans are unlimited.
  • The road to a human is never metered. Neither is the Email the team route.

Every chat is already a ticket​

From the moment you accept the chat notice, the conversation is a support ticket with its own reference of the form OFS-###-####. The first card in the thread — before the welcome, before anything you type — is Your ticket ID, the reference, and "Keep it for future reference — quote it any time you contact us about this."

This is the quiet feature that makes everything else work:

  • Nothing is ever retyped. When you ask for a person, the team opens this conversation — every turn, in order, including what the assistant told you. There is no summary, no second ticket, no "can you explain the problem again".
  • The chat you have open sits in the Continue your chat card on the Home tab, titled with your first question (or Live chat if you have not asked one yet), with a live dot. Every conversation from this browser is listed once under Messages, however many tabs you had it open in.
  • The current chat is remembered in this browser. Continue your chat picks up where you left off; Start new support chat always opens a new conversation.

Nobody is reading along by default. A chat only lands in front of a person when you escalate, or when a teammate chooses to step in.

The composer​

The composer is docked at the bottom of the panel and never scrolls away. With a physical keyboard, Enter sends and Shift+Enter adds a line; on a phone or tablet, the return key adds a line and the send button sends (Ctrl+Enter, or ⌘+Enter, also sends from a keyboard attached to a tablet). Where your account has it, a paperclip sits to the left of the text and you can attach, paste or drop a screenshot for the assistant to read — see Sending a screenshot. Type / for the command palette, and note the delivery ticks under your messages: sending, sent, delivered, and seen once a teammate has read it. A message you were typing survives closing the panel or reloading the page.

Onflow Ads questions only​

The assistant helps with Onflow Ads — your account, campaigns, orders, wallet and how the platform works — and nothing else. An unrelated question or screenshot (another app, a game, homework, a random picture) is declined without being described or answered; on a plan with an AI message allowance it still spends one of those messages, and repeated unrelated use counts as misuse and closes the conversation. A closed conversation — closed by you, by our team, by the idle timer or by misuse — accepts no further messages: start a new chat instead.

After you send, a card shows progress labels such as Reading your message…, Searching the guide… or Looking through your account…, then Writing a reply…. Reading a screenshot, checking a code and building your data report each have their own steps. Then the answer arrives, usually with read-it-yourself buttons under it.

The composer tells you what is expected of you next, by changing its placeholder:

PlaceholderWhat it wants
Type your question…Anything. Normal conversation.
[email protected]Your email address, because you asked for a person. An address you have used before in this browser is pre-filled.
Describe your issue…The one clear message that goes to the team.
6-digit codeThe emailed one-time code inside an account deletion or a data-report request (Plus and above).
Message our team…A teammate has taken this conversation over. You are typing to them.
Type CONFIRM to go ahead, or anything else to leave itThe assistant has proposed a task and is waiting for your word — see What the assistant can do for you.

An End chat button sits under the thread, and the ⋯ menu in the chat header offers Download transcript, Email me a copy, Talk to a person and End chat.

Your allowance​

Every message counts as one — the ones you send and the ones Onflow AI sends back. So an ordinary question and its answer spend two.

Who you areAI messages per cycleWhat that is
Not signed in5Metered against an anonymous browser handle
Free, and Referral Premium30Six times the signed-out allowance, for nothing
Lite, Plus, Elite, Master, UltimateUnlimitedNo meter is drawn at all

The cycle is yours, not the calendar's​

The allowance does not reset for everybody on the first of the month. It runs on a rolling 30-day cycle anchored to your own first message: thirty days later, to the minute, it refills. Two people who started on different days get theirs back on different days — and the widget tells each of them their own exact moment rather than a vague "next month".

The meter​

On a metered plan a pill sits above the composer and counts down live, on every single reply:

● 18 of 30 free AI messages left · back 25 Sep 2026, 2:04 pm — Lite = unlimited

Signed out, the link reads Sign in for 30 instead. On an unlimited plan the pill is not drawn at all. The dot turns amber as you approach the end.

The 75% warning​

At 75% spent, a strip appears above the composer: how much you have used, the exact date and time it resets, and the one step up — "Signing in is free and gives you 30 every 30 days" when you are signed out, or "Lite and above include unlimited AI support" when you are on Free. It stays there until your cycle refills, you sign in, or your plan goes unlimited.

When it runs out​

The panel shows a wall naming the exact moment your own allowance comes back:

  • Signed out — titled "You've used your free AI messages", with Sign in — it's free and See plans.
  • On a metered plan — titled "You're out of AI messages for this cycle", with a button naming the cheapest plan that carries unlimited AI support.
  • Either way, ✉️ Email the team instead sits underneath.

The wall counts honestly. With one message still in hand it does not claim you used all of them — it says you have used n of your limit, and that an answer needs two, your message and the reply.

Make the allowance count

Put the whole question in one message rather than three, and open the buttons under an answer before asking a follow-up. Both halves of an exchange are charged either way, so a well-formed question is worth about two badly-formed ones.

What is never refused, and what is never charged​

RuleWhy
Every step on the road to a person — asking for one, saying yes to an offer, giving your email, describing your issueThese are counted honestly but can never be refused. Being out of allowance must not be able to stand between you and a human.
Anything you send while a teammate owns the threadThat is not an AI turn at all, so the reservation is handed straight back.
An exchange the assistant failed to generateWhen the model produces nothing, the pair goes back onto your meter.
The whole Email the team routeIt has no meter and never has.
The meter gates the AI, never support

There is no state in which you cannot reach a human. Out of allowance, out of turns, engine down, account switched off — the email route is one tap away in every one of them, unmetered, and it reaches the same team.

Asking for a person​

Consent is the rule. Exactly two things start a handover:

  1. You ask for a person in so many words — "talk to a human", "connect me with an agent", "real person", "human help", "escalate", "transfer me", "not a bot".
  2. You reply yes to an offer that is standing.

Nothing else does. A hunch is not consent.

What only produces an offer​

Softer signals put the question on the table and go no further. The offer appears when:

  • the assistant reads you as wanting a person, or as frustrated with AI answers;
  • it could not confidently answer two turns in a row;
  • it could not reach its knowledge base;
  • the conversation passes 40 of your messages.

In every one of those cases your question is answered first, and the offer is appended to that same answer, in one message, so you read the answer before you are asked anything back:

Would you like me to connect you with a human teammate? Reply "yes" and I'll take your details.

It is not repeated turn after turn. Having asked once, it gives you a clean run before asking again. Say no and it steps back: "Alright — I'm here if you change your mind. What else can I help with?"

The three steps​

  1. Your email. "Of course — I'll hand this to a human teammate. First, what email address should their reply go to?" The composer switches to an email keyboard and pre-fills an address you have used before in this browser.
  2. Your issue, in one message. "Got it — replies will go to [email protected]. Now, please describe your issue clearly in one message. I'll pass it to the team together with our whole conversation so you don't have to repeat yourself."
  3. Done. "Done — your request is with our team (Ref OFS-…). They can see our full conversation and will reply right here (and to [email protected]). You can keep asking me questions anytime."

Saying cancel or no at any step returns you to normal chat: "No problem — we'll keep going here. What else can I help with?"

Escalation updates the same ticket — it captures your email, retitles the ticket Escalated from AI chat, reopens it, leaves a note for the desk and sends you the standard acknowledgement email. There is never a second ticket and never a summary lost in translation. And you can carry on asking the assistant questions afterwards; the request lives its own life at the desk.

Make the one clear message count

Include the page you were on, what you expected, what actually happened, and any reference codes. The team reads it with the whole chat attached, but a crisp statement of the problem still saves a round trip.

When a teammate takes over​

A teammate can step into any chat, whether or not you escalated.

  • A system line appears: "👤 Our support team has joined this conversation."
  • Onflow AI goes quiet. The composer placeholder becomes "Message our team…".
  • Their replies render live in the panel within the eight-second poll, marked with their name and · team — and are also emailed to you, subject "Onflow Ads support replied to you [OFS-…]".
  • A team member's name may carry their Admin ID — an OFAM-… reference after the name — so that you can quote exactly who helped you. It identifies them; it is not a way to reach them, so keep writing to support as usual.
  • Your messages during the takeover do not count against your allowance.
  • The assistant stays paused until a teammate resumes it or the conversation closes. Closing clears the flag, so your next chat starts with the AI again.

It is one thread throughout: one reference, every message in order, AI turns and human turns side by side.

When something goes wrong​

What you seeWhat happenedWhat to do
"I'm having trouble reaching my knowledge base right now."The assistant could not answer this turn. The same message carries the offer of a humanReply yes, or try again in a minute
"We've been at this a while — let me get a human teammate to take over."You passed 40 messages in this conversationReply yes, or start a fresh chat for a new subject
The allowance wallYour cycle's AI messages are spentSign in, upgrade, or press ✉️ Email the team instead
"This feature is temporarily unavailable. Please try again shortly."Support could not be reached at all. Nothing was stored and nothing was counted — the reservation is handed backTry again shortly; check the status page
"Couldn't send — please try again." / "Network error — please try again."The request did not completePress the send arrow again
"That doesn't look like a valid email address…"The address you typed during a handover was not usableRetype it, or say cancel to keep chatting
"I couldn't reach the ticket desk just now…"The handover's last step did not landSend your message again in a moment; the conversation is intact
Help-centre answer cards instead of a replyThe assistant is briefly unavailableRead the cards, or press Still need help? Email the team
Support access restrictedThe account or email is barred from the support surfacesEmail [email protected] from the address on the account

Ending, rating, and the transcript email​

  • Pressing End chat closes the conversation and opens the rating screen: How did we do? / "Rate this conversation — it helps us improve."
  • Pick 👍 Helpful or 👎 Not really. A comment box appears — "Anything we should know? (optional)", up to 500 characters — with Send feedback, or press Skip.
  • One rating per conversation; sending again overwrites it. Ratings go to the team as-is and do not change how a request is handled.

Three things close a chat: you pressing End chat, a teammate closing it from the desk, and an automatic sweep that closes any chat with no message for 30 minutes.

Every close emails you a transcript

However a chat closes, you are emailed a copy — subject "Your Onflow Ads support conversation [OFS-…]" — listing each turn as You, Onflow AI or the teammate's name — with their OFAM-… Admin ID, where they carry one — followed by (team). It is sent once per conversation, and skipped entirely when no email address was ever captured, since an anonymous chat has nowhere to send it. If a transcript arrives that you did not ask for, that is why.

Ending a chat clears it from the widget's current-chat memory, so your next question starts a fresh conversation with a fresh reference. The old one is not deleted: it stays under Your conversations, and every email about it links back to it. See Ticket statuses and threads.

Privacy in the chat​

Plainly, because you should not have to guess:

  • Your messages go to a model provider. Each message you send, the conversation so far, and the page you are on are processed by the AI model provider Onflow Ads has configured, in order to generate the reply. This is the same whether you are signed in or signed out — being signed out does not make the chat local.
  • A screenshot you attach goes too, whole. Where your account can send one, the picture is carried to the same model provider with the message — uncropped and unfiltered, so crop it yourself before you send. It is stored with the conversation and deleted in the same operation; one uploaded and never sent is swept within six hours. See Sending a screenshot.
  • The conversation is stored as a support ticket, so the team can read it, and always does when you escalate or step in.
  • On the account-aware plans, a typed summary of your own account, assembled by the server, capped in size and with your identifiers masked, rides along with every message. Only ever yours — never another member's account identifiers, contact details or messages, and never passwords or other sign-in credentials, codes, IP addresses, payout destinations or internal notes. See Answers about your own account.
  • Signed out, the meter counts against your browser, not against you: an opaque random handle in a cookie, with the number of messages used, when the cycle started, and the IP address the last message came from — kept for 180 days after last use, for abuse triage.
  • What never reaches a model provider, in any feature: your password or any other credential, your two-factor secret, your payout destination, and the payment instrument you paid with.
  • You can have AI switched off for your account entirely. Ask at [email protected] or [email protected] and every ticket you open goes straight to a person. Ask before you write about the thing you are worried about — a request typed inside a message has already been sent by the time anyone reads it.
  • Retention. A conversation that was never escalated is deleted 14 days after its last message, once its ticket is closed; the ticket and its messages survive. A conversation you have reported is kept while the team's report is open — otherwise the thing you reported would be gone before anyone could read it — and returns to the ordinary rule once the report is closed.

The full picture is in the Privacy Policy — section 4.6 on what a support conversation holds, and section 12.2 on what reaches a model provider.

Do not paste secrets into the chat

Passwords, one-time codes, card numbers, seed phrases and API keys have no business in a support message, here or anywhere else. Nobody from Onflow Ads will ever ask you for them, and the chat has no use for them.

What each limit means​

RuleDetail
Who can chatEveryone. Signed out: 5 messages a cycle, guide and policy answers only. Free: 30, with answers about your own account. Lite and above: unlimited; actions and flows from Plus.
Message length2,000 characters per message.
Messages per conversation40 from you, after which the assistant offers a person rather than looping.
Conversation memoryRoughly the last twenty exchanges ride along with each question.
Rate limits20 chat messages and 6 new conversations per 10 minutes; a further 40 messages per 24 hours when signed out.
Ticket size200 messages per conversation in total.
AttachmentsScreenshots, on accounts that have them: 3 per message, 4 MB each, PNG/JPG/WebP/GIF — Sending a screenshot. Otherwise paste error text, links and reference codes instead.
Remembered per browserThe current chat lives in this browser. On another device, reopen it from the link in any email about it.
Fair useAbusing the chat or the desk can lead to a support ban: every support surface refuses that account or email, while product access is untouched.
Per-account switchThe team can turn the AI chat off for one account without banning anything. That account's widget goes straight to the guided email flow and a person answers every message.
Last updated on