Technology
Built so your data stays yours.
Ryni is designed around one constraint: a customer's knowledge must never leak to another customer, to a shared model, or to anyone without the right role. Everything else follows from that.
01
AI architecture
Ryni uses retrieval-augmented generation: documents are chunked and indexed per tenant, relevant passages are retrieved for each question, and a language model composes an answer that cites those passages. The model never answers from memory alone.
Retrieval first
Answers are grounded in retrieved passages from the customer's own documents and carry numbered citations.
Model-agnostic
The inference layer is abstracted so models can be swapped or upgraded without changing the product.
Auditable
Every question, retrieved source and answer is attributable to a user and a tenant.
02
Customer isolation
Each customer is a separate tenant from the identity layer down.
One identity realm per customer
Authentication uses OpenID Connect with a dedicated realm per tenant. Users, roles and sessions never cross tenants.
Tenant-scoped storage and indexes
Documents and vector indexes are partitioned by tenant. Every API request is authorised against the tenant in the caller's token.
No cross-tenant caching
The web application never caches or displays data from one tenant in the context of another.
03
Privacy principles
Your data is not training data
Customer documents, prompts and answers are not used to train or fine-tune shared models.
Private inference
Inference runs on infrastructure operated by Ignyr inside the customer's tenant boundary rather than on shared public endpoints.
Least privilege
Role-based access control (for example tenant administrator and analyst) governs who can upload, delete and query.
Minimal collection
We collect what is needed to operate the service and nothing more. This website itself sets no tracking cookies.
04
Cloud infrastructure
Ryni is being built to run on Amazon Web Services, using managed services for compute, storage and networking, with infrastructure defined as code.
AWS-native
Compute, storage, networking and secrets management on AWS, with per-environment isolation between development, staging and production.
Encrypted everywhere
TLS in transit; encryption at rest for documents, indexes and backups.
Observable
Centralised logging and metrics so we can detect, investigate and explain what the platform did.
Ryni is in active development. Infrastructure described here reflects the current design and may evolve before general availability.
Questions about our architecture?
We are happy to walk security and privacy reviewers through the design in detail.
Contact us