Most website AI agents forget the customer as soon as the chat ends. A few remember too much, too quickly. BrainSite is built around a narrower idea: an agent should learn from conversations without letting random visitor claims rewrite the business.
The common problem: agents without a business memory
Plenty of agents can answer from a prompt, a scraped website, or a support article. That is useful, but it is not the same as learning. If every new visitor starts from zero, the agent never gets better at recognizing what people ask, where they get confused, or which answers need more context.
The opposite approach is risky too. If an agent treats every conversation as new truth, a visitor can accidentally or deliberately teach it the wrong thing. A complaint, a competitor's claim, or a half-remembered policy should not become business knowledge for the next person who lands on the site.
BrainSite separates learning from approval
BrainSite learns from two places: approved business context and permitted conversation context. Approved business context is the source of truth. It includes the information the owner adds, reviews, or approves: offers, services, policies, FAQs, website pages, files, corrections, and profile details.
Conversation context has a different job. It helps BrainSite understand what visitors are asking, which gaps keep showing up, and what follow-up questions belong in the flow. It can inform suggestions. It does not silently become approved business knowledge.
Say a visitor tells your BrainSite that a competitor is undercutting your price by 20%, or that your return policy is unfair. Neither statement becomes a fact the agent repeats to the next visitor. A stranger's claim in a chat window is not automatically true, and your business facts should not drift based on whatever someone types at your agent.
What BrainSite can learn
A useful business agent should get better in specific ways. It should notice repeated questions. It should surface missing context for the owner to review. It should keep a visitor conversation coherent when the visitor changes topics, comes back with more detail, or asks a buying question that depends on something they already shared.
That is the difference between memory and drift. Memory helps the agent carry useful context. Drift lets unapproved claims leak into future answers. BrainSite is designed to keep those apart.
How the layers work
| Layer | What it contains | Control | Effect |
|---|---|---|---|
| Approved Knowledge Memory | Business facts, offers, policies, files, and corrections | Owner adds or approves the information | Reusable in future answers |
| Pending knowledge | Suggested facts or corrections | Owner reviews, edits, approves, or rejects | Does not become approved knowledge before review |
| Conversation Memory | Permitted context from an eligible visitor conversation | Separated from approved business facts | Supports continuity within configured limits |
| Website refresh | Proposed changes from a new scan | Changes are staged for owner review | Applied only after approval |
What happens when a source changes?
If the business changes a pricing page, offer, booking rule, or service description, BrainSite can prepare the new context for review. The important part is the review step. A new scan can propose changes, but the owner still controls what becomes live knowledge.
- The scanner collects permitted public website content.
- BrainSite prepares proposed business context, company fields, and sources.
- The proposed changes stay separate from the live deployment.
- The owner approves or rejects the refresh.
- Approved changes become available to the live agent.
Why approval matters
A website agent is often answering questions that affect sales, bookings, pricing, eligibility, refunds, or trust. Those answers need a memory, but they also need custody. The business owner should know what the agent is allowed to rely on.
That is why BrainSite treats approved context as the reusable source of truth, and conversation learning as a way to improve the agent without giving every visitor edit access to the business.
Where owners manage it
Knowledge sources, review controls, memory, and owner workflows live in the docs. Two things stay true regardless of which setting you are configuring: approved knowledge is available for future answers, and unreviewed suggestions stay pending until the owner approves them.
