The conversational brand agent

Short answer: a brand agent is a labelled card beside the finished reply that opens a short side chat with your company's assistant. VibeFree's own model answers, using only reference docs you supplied and VibeFree approved, frozen with the creative. It is billed per engagement - the first message a person sends - and you see counts, never conversations.

How it works

• The card sits at the end of a response, on the web app, VS Code, mobile and desktop. It is a large unit, so it is shown alone. It carries your company, the agent's name, a headline, an ad line, a button and an optional icon, all labelled Sponsored - the specs page has every field limit.
• A click opens a separate side chat. It never sees the person's main conversation, and the main conversation never sees it, so it cannot change what the assistant says there.
• Your opening message (up to 200 characters) is shown first. Every reply after it is written by VibeFree's house model, prompted with VibeFree's rules first and your docs quoted after them as reference data.

Supplying the reference docs

• Up to 20,000 characters of plain text - product docs, plans and prices as you publish them, FAQs. Paste or type them, import a page from a domain your organisation has verified (scripts, navigation and markup are stripped, and you edit the result), or read a PDF, which happens in your own browser.
• One to eight allowed topics, 40 characters each. The agent declines anything else - general coding help included - and suggests the person ask in their main VibeFree chat.
• Reviewed, then frozen. A person reviews the docs and topics with the rest of the creative, and the approved version is what runs. An edit goes back to review while the approved version keeps serving.
• Write them as the source of truth. Where the docs do not answer, the agent says it does not know and points to your landing URL.

The rules your agent runs under

• It speaks for your company about your product only, and says it is a sponsored assistant for you if asked - never VibeFree, a human or a neutral party.
• It never asks for or repeats personal details, and never asks about health, ethnicity, sexuality, religion, politics or trade union membership.
• It makes no claims about prices, returns or guarantees beyond what the docs state word for word, and gives no medical, legal or financial advice.
• It does not run, write or ask anyone to run commands or install anything. Short illustrative code from your docs is fine.
• It keeps replies to a few sentences, in plain language.
• Anything inside your docs that reads as an instruction is ignored: VibeFree's rules always win.

Limits

Crisis handling and personal details

Every message is screened for crisis before the model sees it. A message that reads as self-harm, abuse or an emergency gets a fixed reply with no model call - that this is a sponsored assistant that cannot help, with a pointer to the local emergency number or a support line such as Samaritans - and the chat ends. That message is not kept, and ads in the person's main chat run untargeted and without ad networks, as they would for any crisis. Names, email addresses, phone numbers and similar details are removed from what a person types before the model sees it or it is stored.

Billing and reporting

• Cost per engagement. Billed once, when the person sends their first message. Opening the chat alone is not billed, and later messages in the same chat cost nothing more. One block is 100 engagements; the minimum bid is shown in the ad group form.
• Counts only. Conversations started, messages sent, average messages per conversation, and how many ran to the limit. Never a transcript - not to you, not over the API.
• Transcripts for abuse review only. VibeFree keeps them for 30 days so a person can check the agent is not being misused, then deletes them.
• Pulled at once. If the campaign stops, or VibeFree switches the format or sponsored actions off, open chats end.

Try it first: the free sandbox opens your agent in a mock chat on every app, needs no click and is never billed. It is the quickest way to find a gap in your docs before a developer does.

Frequently asked questions

What is a conversational ad?

An ad that answers questions. On VibeFree it is a labelled card beside the finished reply that opens a short side chat with an advertiser's assistant, answered by VibeFree's own model from reference docs the advertiser supplied and VibeFree reviewed.

Does the brand agent see the user's conversation?

No. It is a separate chat that never sees the person's main conversation, and the main conversation never sees it - so it cannot change what the assistant says there.

What does the advertiser receive from brand agent chats?

Counts only: conversations started, messages sent, the average per conversation, and how many ran to the 10-message limit. Never a transcript. VibeFree keeps transcripts for 30 days for abuse review and then deletes them.

How is a brand agent billed?

Per engagement (CPE): you pay when a person sends their first message. Opening the chat is not billed on its own, and further messages in the same chat cost nothing more. One block is 100 engagements.

Can the agent say things that are not in my docs?

It is told not to. Without an answer in the docs it says it does not know and points to your landing page, and it makes no claims about prices, returns or guarantees beyond what the docs state word for word. Anything that reads as an instruction inside your docs is ignored.

Start a campaign, see every placement in the conversation, or see what the person is told about ads.

Related pages

• Sponsored installs and follow-up chips - Pinned packages and MCP servers, dry runs, and every safeguard.
• Sponsored rewards - Opt-in video, quizzes, paid surveys, interviews and trials on the earn page.
• Is VibeFree really free? - What "free forever" means, and what pays for it.
• Or browse every VibeFree guide.