Key takeaways
- Private knowledge routes require authenticated application access; public marketing pages never expose customer documents.
- Document, user, and retrieval records are scoped to a tenant identifier before application data is returned.
- Security claims on this page describe implemented architecture and provider documentation, not an independent compliance certification.
Security model
SkyDreamKnowledge separates public product information from authenticated company knowledge. Public routes explain the service; private application routes and APIs handle uploaded documents, search, chat, administration, and audit records.
The application uses signed access tokens, server-side membership checks, role-based permissions, and tenant identifiers. An authenticated user must belong to the active tenant before the API provides tenant data. Administrative actions are limited by role.
How company data flows through the system
- Upload: an authenticated user submits an approved document to the tenant knowledge base.
- Store: the original file is stored in a private Cloudflare R2 bucket using a tenant-scoped object key.
- Index: document content and source metadata are prepared for Cloudflare AI Search.
- Retrieve: the server sends a tenant-scoped query and receives relevant passages.
- Generate: retrieved context is supplied to the configured AI model to produce a grounded response.
- Return: the user receives the answer plus source metadata for authenticated preview.
Cloudflare’s current AI Search architecture documentation describes ingestion, chunking, embedding, vector and keyword indexing, retrieval, reranking, and optional response generation.
Tenant isolation and authorization
A tenant represents a company workspace. User membership, documents, bots, usage records, and audit events are associated with a tenant identifier. API middleware verifies the signed identity and confirms that the stored membership still matches the token before protected operations continue.
Tenant identifiers are also included in document storage and retrieval metadata. This is essential: interface-level separation alone is not sufficient for a multi-tenant RAG service. The server must enforce the boundary before returning documents or retrieved passages.
Authenticated source access
Citation links use application-controlled document routes rather than publishing customer files as crawlable assets. The same authenticated context used for search is applied when a user opens a source preview.
Cloudflare services used
- Cloudflare Workers: runs the Nuxt application and protected API logic.
- Cloudflare D1: stores relational metadata such as tenants, users, subscriptions, and audit records.
- Cloudflare R2: stores uploaded document objects.
- Cloudflare AI Search and Workers AI: support indexing, retrieval, reranking, and response generation.
Cloudflare documents that R2 objects and metadata are encrypted at rest and transferred using TLS. Cloudflare also states in its Workers AI data usage policy that customer content is not used to train Workers AI models or improve Cloudflare or third-party services without explicit consent.
Data-location requirements need deliberate configuration. Cloudflare documents jurisdiction controls for services such as D1; the correct setup depends on the customer’s contractual and regulatory requirements.
Customer and administrator controls
- Role-based user administration within each tenant.
- Document upload, indexing status, preview, and deletion.
- Audit records for important knowledge-base and administrative activity.
- Configurable RAG behavior and retrieval settings for authorized plans.
- Separate credentials and validation for Slack, Mattermost, and email channels.
- Tenant-scoped subscription and usage controls.
Scope, assurance, and limitations
This page is an architecture overview, not a penetration-test report, legal opinion, data processing agreement, or claim of SOC 2 certification. Cloudflare’s certifications do not automatically certify applications built on its services.
Organizations should evaluate SkyDreamKnowledge against their own data classification, identity, retention, residency, incident-response, vendor-management, and compliance requirements before uploading sensitive material. Access-control and deletion behavior should be tested in the intended deployment environment.
For questions about a specific deployment or data-processing requirement, contact support@skydreamcity.online before production use.