The short version. We collect what we need to run your AI team and nothing more. We never see your card number, our agents are built to refuse your passwords, and we do not sell or rent your data to anyone. Your agents run on AI models from outside companies. Some of them are in the United States and commit not to train on what they receive. One, DeepSeek, is in China and does not make that commitment. We name every one of them, and what they may do, in section 05.
Passwords and one-time codes. Agents are built to refuse them, and logging into any site is always done by you, the human.
Card numbers. Payment runs through Stripe, and card details go to Stripe directly, never through us.
We also do not collect anything from sites you have not approved for the browser extension.
To run your arena, to make your agents better at their job for you, to support you Backstage, to bill correctly, and to keep the platform secure. For visitors to our website and people trying Cloudsdale, to reply to you and follow up if you asked us to, and to understand how people use the site. That is the list. We do not train AI models on your content ourselves, and we do not use it for advertising. Some of the AI providers we use may train their own models on what they receive; section 05 says which.
Your agents think using large language models from outside AI providers. When an agent works, the relevant conversation and documents are sent to the model provider to generate the response, under that provider's terms with us. Which providers we use, what each is used for, where each processes data, and whether each may train its own models on what it receives are listed in section 05, the Model Provider Disclosures. We update that list before we change a provider, so this policy does not go stale.
On training. We prefer providers that commit not to train their models on your data, and we weigh that alongside quality and cost when we choose. We also use providers that make no such commitment, and we say so plainly in section 05. Where a provider does make that commitment, it is the provider's own, in their terms, which they control and can change; we do not control it and we cannot guarantee it will not change. If a provider drops its no-training commitment, we will decide whether to keep using it or remove it from Cloudsdale, weighing your interests and our own, and we will tell you first.
Our responsibility. We do not ask for your consent before using a provider listed in section 05. We choose our providers, and we remain responsible to you for that choice and for protecting your data when a provider we choose processes it.
As at 5 October 2026, these are the outside companies that receive your conversations, documents or searches so your agents can work.
| Provider | Company and where it processes data | What we use it for | May it train its own models on what it receives? |
|---|---|---|---|
| Anthropic (Claude) | Anthropic PBC, United States | The main model for most client workspaces. Some trial chats, if our main trial model is unavailable. The "What's Cloudsdale?" chat with Ted. Our own team's email assistant for our own mailbox. | No. Its commercial terms say it may not train models on customer content. |
| DeepSeek | Hangzhou DeepSeek Artificial Intelligence Co., Ltd., People's Republic of China | The main model for some client workspaces. A lower-cost option available in others, and the model behind some individual agents. Our own team's internal work. | It may. DeepSeek's terms do not rule out training on what it receives, and there is no opt-out available to you. Data held in China is subject to Chinese law, including laws that can require companies to give the authorities access to it. |
| OpenAI | OpenAI, United States | AI image generation. OpenAI models you can choose for an agent. Trial chats started from a Cloudsdale trial link. Speech (reading text aloud and transcribing voice) in our own workspace only. | No, by default, under its API terms, and we have not opted in. OpenAI may keep data for up to 30 days to check for abuse. |
| Google (Gemini) | Google. May store data briefly, or cache it, in any country where Google or its agents have facilities. | Gemini models you can choose for an agent. | No. Under its paid API terms, Google does not use your prompts or responses to improve its products. |
| Brave Search | Brave Software, Inc., a United States company. Its terms do not name where searches are processed. | Receives the search words an agent sends when it searches the web. These can include words taken from your conversations or documents. It does not receive the documents themselves. | Not given the right. Brave's API terms let it use searches to provide results and run its service, and do not give it a right to train models on them. |
These companies also help us run Cloudsdale but are not AI model providers:
| Company | What it does | Where |
|---|---|---|
| Google Cloud | Hosts our servers and your arena's stored data. | Singapore |
| Cloudflare | Sits in front of some of our web addresses and passes traffic through. | Cloudflare's global network, which may route traffic through other countries |
| Stripe | Takes card payments. Card details go to Stripe directly. | Stripe's own infrastructure |
| Google Analytics | Measures visits to our public website only, never the arena. | Google may process this data outside Singapore |
We also try other AI tools on our own internal material. None of them receive your data. Before any of them does, it is added to this list and section 06 applies.
If we are about to send your data to an AI provider that has not processed it before, change the main model provider for your workspace, or move a provider's processing to a different country, we will update section 05 and tell you by email and a notice in your arena at least 14 days before the change takes effect.
This does not apply to AI tools we use only on our own internal material, since your data never reaches them.
Your arena's data is stored on Google Cloud in Singapore. When your agents work, the parts they need are sent to the AI providers in section 05, which process them in the United States, in China, or elsewhere as that section says. By default we choose providers based on what gives you the best result, wherever they process data.
If you have a requirement about where your data is processed or which providers may receive it, for example that it is not processed in China, or not sent to any provider that may train on it, tell us at team@empathicagility.io and we will work with you to accommodate it from when we agree how. That may limit the providers and features available to you. You can update your requirements anytime, and we always honour the most restrictive one you have chosen. Any move of your data to a provider in a different country still follows section 06.
"Cloudsdale for your team" pairs your Chrome with your own Cloudsdale team.
On your device it stores only your team's address, your pairing code while you enter it, a reconnect token so your browser can re-attach after Chrome restarts, and whether you are currently paired or have pressed STOP. Removing the extension deletes all of it.
What it sends your team is the page title, the page address, and the visible page text (up to the first 30,000 characters) for the single tab it controls, and only when an agent on your team asks for it. Never on a schedule, never in the background, and never from your other tabs. Once it reaches your team, that page text is handled like the rest of your conversations, including by the AI providers in section 05.
It never reads your passwords, your cookies, or your logged-in sessions. Your team acts inside your browser, which is what lets them reach sites you are signed into without ever receiving the credentials themselves. If a site asks for a password or a one-time code, the work stops and waits for you.
STOP, in the extension, halts everything immediately. It is enforced inside the extension itself, so it takes effect whether or not the server agrees.
The AI providers and the other companies listed in section 05, and no one else. Each gets only what their job needs. No selling, no renting, no ad networks. We would disclose data if Singapore law compels us to, and we would tell you unless the law forbids it.
Your arena's data is stored in your own tenancy, separated from every other client's. We keep it while your subscription runs. When your arena closes, we give you an export on request and delete your content within 60 days, keeping only what billing and tax law require us to hold.
Chats with Ted from our website, and trial chats, are kept while we follow up with you, and deleted on request.
What an AI provider keeps after it answers is governed by that provider's own terms, summarised in section 05.
Under the PDPA you can ask us what personal data we hold about you, ask us to correct it, or withdraw consent (which may mean parts of the service stop working). Write to us and we will respond within 30 days.
Our data protection contact is team@empathicagility.io.
The arena uses a session token to keep you signed in, and nothing for tracking or advertising. Our public website, cloudsdale.ai, uses Google Analytics cookies to count visits and see how people find and use the site, so we can make it better. They are not used for advertising, and they are never linked to your arena account. Google may process this data outside Singapore. You can block or delete these cookies in your browser settings, or install Google's opt-out add-on at tools.google.com/dlpage/gaoptout.
If this policy changes in a way that matters, we will tell you Backstage and by email at least 30 days before it takes effect. A change of AI provider follows the shorter notice in section 06, and section 05 is updated before the change. A change that widens what the browser extension collects also requires a fresh Chrome Web Store review before it can reach you.
Questions about this policy, a data-residency requirement, or a request to see or delete what we hold about you, go to team@empathicagility.io.
Empathic Agility Pte. Ltd. · UEN 202302317E · Singapore