Trust and security
Written so you can audit the claims. Where something is planned and not shipped, this page says so, because a gap that is hidden costs more trust than the gap itself.
The short version
Ownership. You own your data, you can export it in bulk at any time, and you can delete it. We do not train public models on it.
Where it lives. Google Cloud Platform, in the United States, with standard encryption in transit and at rest.
Which AI. Calls are routed to Google's Gemini models, chosen per job, such as writing, embeddings and image generation. Google is contracted for no-training commitments on your data.
Current limits. Incident response is slower outside business hours, and formal compliance documentation is thin. Both are covered below.
Not enterprise-ready. No SOC 2, no SSO on the base plan, no DPA as a default. All on the roadmap. Ask if you need specifics today.
Subprocessors
The vendors your data passes through to run the product. A list you can read is more useful than one you have to request.
| Subprocessor | Purpose | Data touched |
|---|---|---|
Google Cloud Platform | Primary infrastructure, compute, storage, database, analytics (Cloud Run, Cloud SQL, Cloud Storage, BigQuery, Cloud DNS, Memorystore). | All customer data. |
Firebase (Google) | Sign-in. | Account identity. |
Google AI / Vertex AI | AI generation (Gemini models), embeddings, image generation. | Prompts, completions, embedding text, image generation requests. |
Resend | Transactional and marketing email delivery. | Recipient email addresses, email subject and body, delivery and engagement events. |
Stripe | Billing and payments. | Payment method, billing contact, invoice data. |
What we do with your data
You own it
Pages, posts, emails, drafts, KB articles, CRM records, and the context you upload are yours. Export everything in bulk at any time. If you leave, you leave with your data.
No training on your data
Google, the provider we route to, is contracted under terms that prohibit training on customer data. We do not maintain our own foundation model or expose your content to another customer.
Voice via context, not fine-tuning
Your voice lives in retrievable context the AI pulls at generation time. It is not baked into shared weights.
We do not sell data
Not to advertisers, not to data brokers, not to anyone. We keep operation logs scoped to your organization for audit and billing, and we do not mine them across customers. The meter is the only thing we sell.
Access, encryption and where the data lives
Your organization owns its projects, and only members of your organization can see your content. Each member holds one role in the organization and a separate level of access to every project and every app inside it, so a person reaches what you gave them and nothing else. An API key belongs to one organization and reaches only that organization's data.
Anything acting on your behalf carries the permissions of the account it belongs to, whether that is an API key, the SDK or an AI client connected over MCP. Change the account's permissions and you change what it can reach.
Internally, production data access is logged. We do not read your content, inspect your drafts or browse your CRM as routine. Production access is held by a single administrator, used sparingly and logged. Internal access tiers become more formal as the company grows.
Encryption is TLS 1.2 or higher for every request, with public endpoints enforcing HTTPS, and GCP-managed AES-256 at rest for Cloud SQL, Cloud Storage and BigQuery. Data residency is the United States today. EU residency is a roadmap enterprise option and not available now. If you require EU residency today, we are not your tool yet.
Backups run daily across all Cloud SQL databases, with point-in-time recovery inside a 30-day window and per-project JSON and CSV exports on demand. When you delete a project, its content goes to a trash for 30 days and is then hard-deleted. When you delete an organization, all project data and customer records are wiped, with hard deletion from active systems within 30 days and removal from backup rotation within the backup window. There is no clause that retains your data past deletion.
Compliance
What exists, what is planned and what is not. Open items are listed as open rather than given a date.
Not yet. SOC 2 Type I, then Type II, is the first formal compliance program and the priority for the company through the next stretch. The audit process is to begin in the near term, with no date committed. ISO 27001 is planned after SOC 2 Type II.
Not on the base plan today. SSO and SAML are available on request as a flat-priced enterprise add-on, with self-service provisioning on the roadmap. Email plus password, Google, and Microsoft sign-in are supported, with two-factor authentication available.
Not today. Data lives in US GCP regions. EU residency with EU-only data pinning is a planned enterprise option, not shipped. Regional options beyond US and EU are not supported.
No HIPAA and no BAA. We do not support healthcare workloads. No FedRAMP and no government workloads. Payment data is handled entirely by Stripe. We do not touch card numbers directly, so our own PCI scope is minimal by design.
Error monitoring and alerting run on all services, with on-call coverage. A confirmed security incident affecting customer data means direct email to affected customers and a written post-mortem after resolution. There is no formal SLA today. The target is first response within one hour during business hours, and best-effort outside them. Response is slower overnight. If you need tighter commitments today, we are not your tool yet.
Email security@giantcontext.com before posting publicly. We acknowledge receipt within two business days, investigate, and fix on a timeline proportional to severity. We'll credit you by name in the post-mortem if you want credit. There is no formal bug bounty program today, but we'll work out something fair case by case.
Something this page did not cover
For a signed subprocessor list, a DPA, a contractual RTO or RPO, or anything a purchase decision hinges on, write to security@giantcontext.com. If we cannot commit to it, we will say so.