Data use
- What holds it
- Never trains a model, never shared across customers; knowledge bases bind to named workers, so visibility is configured, not global
- What you can inspect
- The clause · the bindingPrivacy §4
AI talks to customers. Where does the data go?
Three things up front: customer conversations never train a model; each region keeps its data in its own deployment; anything said in your name is still yours to answer for, so the red lines, guard rules, human takeover and replay stay in your hands.
Three questions, answered
None of this was written for this page. It is already in our Terms of Service and Privacy Policy.
Will it train your models?
No. Your data is not used to train or improve any machine learning model, and we do not pass it to model providers for their training. The review workbench runs on your own cases; nothing is shared across customers.
Terms §5 · Privacy §4
Terms §5Privacy §4Must I tell customers it is AI?
It cannot deny it. Platform rules forbid an agent from impersonating a specific real person, or from denying it is an AI when an End User asks. Whether you must disclose upfront depends on your jurisdiction.
Terms §7
Terms §7Who answers for mistakes?
Anything sent in your name is yours to answer for. You are the controller; we process on your instructions as a processor. So we give you the mechanisms that lower the risk, not a promise that nothing will.
Terms §6, §8 · Privacy §1
Terms §6, §8Privacy §1Compliance you can inspect, not just be promised
Anyone can write a promise. Each of these six rows points at something real in the product — open them one by one during the demo.
Location
- What holds it
- Two independent deployments; data stays in the region of the site you signed up on — no cross-border transfer
- What you can inspect
- Privacy §7Privacy §7
Red lines
- What holds it
- What it must say, sensitive-topic limits, forbidden tools: assertions a machine checks line by line. No pass, no launch
- What you can inspect
- Assertion resultshow it is proven safe before launch
Limits, at runtime
- What holds it
- Every stage has a goal and entry conditions; guard rules decide whether it may move on — not the model’s judgement
- What you can inspect
- Stage defs · guard ruleshow the playbook is enforced
If it oversteps
- What holds it
- Switch any contact between human and AI; the handoff is backstage, and messages can be reviewed first
- What you can inspect
- Who owns each contacthow human takeover works
Tracing it back
- What holds it
- Replay every turn — reasoning, tool calls, latency; configuration is versioned, so you can roll back
- What you can inspect
- Trace replay · versionshow to find which step failed
What is yours, and what is ours
We did not pick this split. The law did: for your End Users’ personal information you are the controller, and we process on your instructions as a processor.
You · the controller
Lawful basis / Notice & rights / Outgoing
- Before contacting an End User, get consent or confirm another lawful ground
- Give the required notices; access, correction and deletion requests come to you
- Anything sent in your name is yours to answer for; you draw the review and takeover lines
Us · the processor
Us · the processor
- We neither obtain nor judge it for you. No platform can take this layer
- We help you answer on your instructions, never handling a request behind your back
- We provide review, guard rules, takeover and replay to lower the risk. That is not a warranty of accuracy
Where the data sits, and if it leaves
Two sites, two independent deployments. Your data is processed and stored in the region matching the site you signed up on.
Mainland China
topppai.cn serves mainland China, and its data is stored in mainland China.
Outside China
toppp.ai serves users outside mainland China, and its data is stored outside mainland China.
The two never meet
No personal information crosses the border. If this arrangement changes, we will tell you in advance and follow the procedure the Personal Information Protection Law requires.
Tenant and project are two layers of scope. Your knowledge, material, test cases and conversations sit in your own workspace, shared with no other customer. Privacy Policy §4 · §7
Telling customers they are talking to AI
Two hard rules in the platform terms, plus the one that gets misread most.
No impersonation
An agent may not be configured to pose as a specific, real individual. It can hold a role and a name. It cannot be someone who exists.
No denying it is AI
If an End User asks whether they are talking to a bot, the agent may not deny that it is an AI. That is an acceptable-use rule, not a setting.
Takeover is not hiding
When a person takes over, it happens backstage and the customer keeps the same advisor. That protects continuity of service. It is not there to hide what the other side is. You can always see who holds a contact.
Whether you must disclose at the start of a conversation varies by jurisdiction. Yours governs.
Three cases where we do not fit
Better said here than discovered halfway through a demo.
You want to outsource liability
You cannot. For your End Users’ personal information the controller is you, and that role does not transfer to a platform. We give you mechanisms, not cover.
You blast without a lawful basis
Platform rules forbid harassing, high-frequency or indiscriminate outreach, and contacting anyone who has asked you to stop.
You only want a bot that answers and stops
Cheaper options exist. What custom buys you sits in follow-up and closing; on answering alone it never shows, and the price feels wrong.
Five questions to ask before you pick anyone
They apply to us too. Copy them into your evaluation sheet and ask every vendor.
01 · Use
Will conversation data train a model? Get it in the contract, not in a sales call.
02 · Location
Which region holds the data, and are the deployments truly independent?
03 · Red lines
How is it proven safe before launch? Can you see per-assertion results, not one score?
04 · Limits
What stops an over-authorised promise — wording in a prompt, or an entry rule?
05 · Trace
Can you find which step failed, and confirm which configuration version was live?
Four more things people ask
Will our conversations be used to train your models?
No. TOPPP does not use Customer Data to train or improve any machine learning model, and does not pass it to model providers for their training. If that ever changes, TOPPP will tell you prominently before the change takes effect.
Can I have data deleted?
Yes. When you delete content, TOPPP removes it from production, and backup copies are cleared on the backup rotation cycle. After you close the account, account information is deleted or anonymised unless the law requires it kept.
Will my data be used to serve another customer?
No. TOPPP does not use your Customer Data to provide the Service to anyone other than you, and the review workbench runs on your own test cases and conversation records. You can ask TOPPP for the list of processors serving your account.
Who handles End User rights requests?
You. For End Users’ personal information you are the controller, so requests go to you; TOPPP acts as the processor and helps you answer on your instructions.
Bring the checklist. We will go down it
A demo runs about 30 minutes. The compliance half goes like this:
- 01Walk the five questions, showing the mechanism for each
- 02Play a customer in your industry, talking to an AI sales worker
- 03Watch the agents hand off and reason, live, on the right
Contact business@toppp.ai