01Who we are
DocsGPT Cloud is operated by ARC53 LTD, a company registered in Scotland under company number SC642841, with its registered office at 5 South Charlotte Street, Edinburgh, Scotland, EH2 4AN, United Kingdom (“Arc53”, “we”, “us”).
- Privacy questions and rights requests: privacy@docsgpt.cloud
- EU representative (GDPR Art. 27): contact our representative at privacy@docsgpt.cloud
This policy explains what we do with personal data when you use DocsGPT Cloud (the “Service”). It does not create consent by itself: where we rely on consent, we ask for it separately.
Where you run DocsGPT on your own infrastructure, Arc53 processes none of your data. Where Arc53 operates a dedicated instance for you, the agreement for that instance applies in place of this policy.
02Our two roles
We are the controller — we decide how the data is used — when you sign up as an individual, for your account, billing, support, security and the Service itself.
We are a processor — we act on someone else’s instructions — when a business customer uses the Service and we handle personal data on their behalf. That includes content that a customer’s team uploads, and data from people who interact with an Agent that a customer deploys. In that case the customer’s own privacy notice governs the data, our Data Processing Addendum governs our handling of it, and you should send rights requests to that customer. We will forward any we receive.
03What we collect
3.1Information you give us
- Account: Name, email address, profile image, password or social sign-in, organisation and team membership, roles. Handled by our authentication provider, Clerk.
- Billing: Plan, subscription status, billing contact, and the payment details you give Stripe. We do not see or store full card numbers.
- Content: Prompts and conversation messages; uploaded files and attachments; sources you ingest from links, repositories or connected accounts; the text, chunks, vector embeddings and knowledge-graph entries derived from them; memories, notes and to-dos; agent, prompt, workflow and schedule configuration; generated artefacts and images.
- Voice: Audio you record for transcription, and text you send for read-aloud.
- Integration credentials: API keys and OAuth tokens you add to connect other services, and data those services return.
- Support: Messages, feedback, thumbs up/down ratings, and bug reports.
3.2Information we collect automatically
- Usage: Features used, conversations and messages created, models selected, tool calls made and whether they succeeded, token counts, and session activity.
- Technical: IP address and request metadata at our network edge, browser and device type, operating system, language, and identifiers for devices you pair with the Service.
- Diagnostics: Error and performance logs. These are designed to record identifiers rather than content, but some entries include the text of a failing query or the name of a file.
- Cookies: See section 10.
3.3Information from third parties
- Clerk, and Google if you sign in with it: your identity and profile.
- Connected services you authorise, such as Google Drive or GitHub: the content and metadata you point us at.
- Company information. When someone signs up with a work email, we look up public information about the organisation at that domain, using People Data Labs, so our team knows who is trying the product. We send the domain, not your email address.
04How we use personal data, and our legal basis
| Purpose | Basis (UK/EU GDPR) |
|---|---|
| Providing the Service: answering your questions, ingesting and searching your sources, running agents, workflows, schedules and tools | Performance of a contract |
| Accounts, authentication and organisation management | Performance of a contract |
| Billing, invoicing, collections, and tax records | Contract; legal obligation |
| Support and service communications | Contract; legitimate interests (helping you use the product) |
| Security, abuse prevention, rate limiting and fraud detection | Legitimate interests (protecting the Service and its users); legal obligation |
| Diagnosing faults, and reproducing and fixing them (section 6) | Legitimate interests (a working, reliable product) |
| Aggregate product analytics and service quality measurement | Legitimate interests (understanding what to improve) |
| Human review of how the Service handled a request, to correct retrieval, parsing, tool selection and safety (section 5.4) | Legitimate interests (an accurate, reliable product), which you can object to at any time |
| Marketing emails about our products | Consent, or legitimate interests for existing customers, with an unsubscribe link in each message |
| Complying with the law and handling legal claims | Legal obligation; legitimate interests |
Where we rely on legitimate interests, we have weighed them against your rights, and you can object at any time (section 9).
05AI models: what we do and don’t do with your data
5.1We do not train AI models on your content
Nothing you put into DocsGPT Cloud — prompts, answers, uploaded files or connected sources — is used to train, fine-tune or otherwise adjust the weights or parameters of any AI model, ours or anyone else’s, and none of it is placed in a training, fine-tuning or evaluation dataset. This applies to every account on every plan. There is no setting that changes it and no programme you can be enrolled in.
By “AI model” we mean any machine learning model: the large language models that write the answers, and the models used inside the Service for retrieval and ranking, document parsing, tool selection, routing and guardrails.
This promise is about content. We also record how the Service behaved — which tools ran, what failed, how long things took, how a document parsed — and we use that to make the product work better, as described in 5.4 and 5.5.
5.2Model providers
To answer your questions, we send prompts, the relevant parts of your sources, attachments and tool results to the model provider you or your agent selects (section 7). We use those providers on terms that prohibit them from training on the data we send. Some of them keep a copy for a short period — typically up to 30 days — for their own abuse monitoring, and where a conversation is continued through a provider’s stateful API, the earlier turns are held by that provider for the same kind of period.
5.3Files, and things that break
Files reveal problems that nothing else does: a PDF that will not parse, a spreadsheet that produces nonsense, an encoding that crashes a converter. When your file triggers an error we may keep a copy of it, and of what we extracted from it, in restricted storage so we can reproduce and fix the problem. Only named engineers can open it, it is used for nothing else, and it is deleted within 30 days, or sooner if you ask.
To stop the fault returning we want a permanent test case. We build one that keeps the structure of the file but replaces its content with filler, so the test contains none of your data. We do not keep your file itself as a test case.
5.4Human review, and improving how the Service works
Some things can only be judged by looking. To tell whether the Service answered well, chose the right tool, parsed a document correctly, retrieved the right passage or triggered a safety rule when it should have, our staff may review the material involved — the prompt, the answer, the source document and the tool calls — and use what they learn to correct how the Service behaves: its retrieval and ranking, parsing, tool selection, routing and guardrails.
What we learn is applied by changing rules, thresholds, configuration and the instructions we give the models. It does not put your content into any model’s training data, and it never touches the API keys, OAuth tokens or other credentials you have stored.
Review is limited to named staff, access is logged, and everyone involved is under a confidentiality obligation. You can switch it off at any time, at no charge, for your account or your whole organisation, by emailing privacy@docsgpt.cloud. Where we act as a processor for a business customer, this review is part of providing and maintaining the Service under that customer’s agreement, and the DPA governs it.
5.5Aggregated and de-identified statistics
We produce statistics about how the Service performs and how it is used — for example how often a tool is chosen, how long answers take, or how often a document type fails to parse. These are aggregated and de-identified so they do not identify you, your organisation or your content. They are not personal data, we own them, and we may keep, publish and use them without restriction, including to operate, improve and develop our products.
5.6Zero Retention Mode (Enterprise add-on)
Enterprise customers can purchase Zero Retention Mode. Where it is switched on for a workspace, we turn off provider-side storage of conversations and files wherever the provider offers that for the deployment in use, keep content out of our application logs, and shorten or remove the fault-retention in 5.3. It is not switched on by default, and it is not included in the standard Enterprise plan — contact contact@docsgpt.cloud. You do not need Zero Retention Mode to switch off human review: 5.4 explains how.
06Diagnostics and support access
A small number of named Arc53 staff can access service logs, and, where it is needed to investigate a specific fault or a support request, the affected content. Access is limited to what the investigation needs, is logged, and is subject to confidentiality obligations. We do not browse conversations.
08International transfers
Arc53 is in the United Kingdom, and our main systems are in the UK and the EEA. Some providers in section 7 are in the United States.
Where data leaves the UK or the EEA for a country without an adequacy decision, we rely on the UK Extension to the EU-US Data Privacy Framework, or on the standard contractual clauses with the UK Addendum or the UK International Data Transfer Agreement, together with an assessment of the destination country’s laws. Personal data reaching us from the EEA relies on the European Commission’s adequacy decision for the United Kingdom.
Ask us at privacy@docsgpt.cloud for details of the safeguards for any specific transfer.
09Your rights
Under UK and EU data protection law you can ask us to:
- give you a copy of your personal data, and provide it in a portable format;
- correct data that is wrong or incomplete;
- delete your data — you can delete conversations, sources and agents yourself at any time, and delete your whole account in Settings;
- restrict or object to processing based on legitimate interests, including the human review described in section 5.4 — email us and we will switch it off;
- withdraw consent where we relied on it, such as for marketing emails, without affecting what we did beforehand.
Email privacy@docsgpt.cloud. We answer within one month, and will tell you if we need longer. We may ask you to confirm your identity. There is no charge unless a request is excessive.
If we act as a processor for a business customer (section 2), send the request to them.
Complaints. You can complain to the UK Information Commissioner’s Office (ico.org.uk) or, in the EEA, to your local supervisory authority. We would rather hear from you first.
11How long we keep data
| Data | Kept for |
|---|---|
| Conversations, sources, agents and other content | Until you delete it, or your account closes |
| Account and profile | While the account is open |
| Billing records and invoices | 7 years, for tax and accounting law |
| Streaming message journal | 14 days |
| Safety and guardrail events | 30 days |
| Scheduled run outputs | 90 days |
| Code sandboxes | Stopped after 15 minutes idle, deleted an hour later |
| Files kept to reproduce a fault (section 5.3) | 30 days |
| Application logs | 90 days (Axiom); error records 90 days and log streams 3 days (Better Stack) |
| Backups | Deleted data disappears as the backup window passes, within 35 days |
| Inactive free accounts | Deleted after 12 months, with notice first |
When you close your account we delete your data within 30 days, apart from backups and logs within the windows above, and anything we must keep by law.
12Security
We protect personal data with measures including: encryption in transit and at rest; encryption of the third-party credentials you store; access control, roles and single sign-on; tenant separation; isolated, short-lived sandboxes for code execution; least-privilege production access by named staff; centralised logging and monitoring; code review and automated testing; and vendor due diligence.
No system is completely secure. If a breach affects your personal data and is likely to present a risk to you, we will notify the Information Commissioner’s Office within 72 hours where required, and tell affected users and customers without undue delay.
13Children
The Service is not for children. You must be at least 16 to use it. We do not knowingly collect data from children below that age, and we delete it if we find it. Contact privacy@docsgpt.cloud if you think a child has given us data.
14Automated decisions and AI
The Service generates answers automatically, and agents can take actions you have configured. We do not use it to make decisions about you that produce legal or similarly significant effects without human involvement.
AI output can be wrong: check anything you rely on. Where you deploy an agent to other people, you are responsible for telling them they are dealing with an AI system.
15United States privacy rights
If you live in California or another US state with a privacy law, you may have the right to know what we collect, to access, correct, delete or port it, to opt out of sale, sharing or targeted advertising, and to be free from discrimination for exercising those rights.
We do not sell personal information, and we do not share it for cross-context behavioural advertising. To exercise a right, or to have an authorised agent act for you, email privacy@docsgpt.cloud. We will verify the request and answer within the statutory time.
16Changes to this policy
We may update this policy. For material changes we will give notice by email or in the Service at least 30 days beforehand, and we will update the date at the top. Earlier versions are available on request.
17Contact
ARC53 LTD (company number SC642841), 5 South Charlotte Street, Edinburgh, Scotland, EH2 4AN
- Privacy: privacy@docsgpt.cloud
- General: contact@docsgpt.cloud
- EU representative (GDPR Art. 27): contact our representative at privacy@docsgpt.cloud
We are not required to appoint a Data Protection Officer and have not done so.