Noezra Provider Register
Noezra uses the providers below to operate storage, AI, speech, analytics, and research discovery features. Private Workspace Content is not used for shared model training at launch.
AWS is primarily configured in Sydney (ap-southeast-2). PDF Inspector performs local document processing first and only pages needing OCR use self-hosted MinerU 3.4 when safe. If selected-page OCR or composition cannot safely complete, whole-document MinerU is used before any eligible emergency Bedrock Data Automation decision. BDA is never the primary, default, or normal extraction route; it is only an emergency fallback using an APAC cross-region profile. Provider locations and retention depend on the active configuration and contract; this register does not promise exclusive residency or zero retention. Anthropic is no longer used for new AI processing. Earlier processing remains subject to the applicable provider terms and retention requirements; model retirement does not itself delete those records.
Hosted subscription checkout, payment-method management, invoices, renewal, cancellation, and configured tax calculation. Stripe receives billing details and subscription references; private research and workspace content is not sent for payment processing.
Sydney-primary hosting, database, storage, queues, identity, email, monitoring, and security. Local PDF Inspector processes PDFs first, self-hosted MinerU 3.4 performs normal OCR in Sydney, Amazon Bedrock Titan Embeddings V2 processes embeddings in the existing AWS production region, and Bedrock Data Automation is only a budget-controlled emergency APAC fallback.
OpenAI chat, grounded answers, summaries, and generated titles in the United States or globally only for the requested AI feature.
Chat inference through Microsoft Foundry in the United States data zone or globally, depending on the deployment, only for the requested AI feature.
Public DNS management and, where proxying is enabled, edge delivery, traffic protection, and security.
Google Sign-In through Amazon Cognito when selected by the user.
Text-to-speech audio generation and browser delivery of ElevenLabs-hosted public-site assets.
User-requested dictionary definitions and pronunciation lookup; lookup terms and browser request metadata are sent.
Optional pseudonymous product analytics.
Web-font delivery for the Noezra interface.
Public topics and verified public paper metadata for requested searches; bounded, minimized concepts from available authorised chat sources when you request a paper search. Uploaded passages and private notes are not sent to search. Leads are verified against scholarly indexes before display. Bing processing is separate from Foundry model processing, may occur outside its geography boundary, and uses Microsoft Bing terms.
Research discovery and public paper metadata lookup.