Architecture
Nile is taking a radical approach to database platforms. The goal is to be the best platform for building AI-native B2B companies. Our core insight is that it is possible to push tenant/customer awareness all the way to the page level in Postgres and design it from first principles for the cloud and an AI-powered future.
Nile's architecture includes storage/compute decoupling, page-level tenant isolation, a global routing gateway, and a database clustering technology to provide the following key benefits -
- Built-in tenant virtualization in Postgres for better data and vector embedding isolation and performance isolation without any RLS or complex permission logic in the application across customers.
- Vector embedding with RAG support for all model providers, cost-effective and auto-scales to billions of embeddings across customers
- Use one database in your application but get tenant-level backups, customer/tenant insights, and schema management across tenants managed by Nile
- Seamlessly auto scales as your customer's usage increases and scales to zero with no cold start time
- Place some tenants/customers on dedicated compute in the same database and some in other regions. Nile lets you customize for cost, security, compliance, and latency for each tenant in your AI-native application without having to manage separate databases for each tenant
- You can scale to millions of customers and billions of vector embeddings effortlessly as Nile horizontally scales
- User management is built for multitenancy with user data stored in your Postgres database and unlimited active tenants and users.
We will have a more detailed write-up about our blueprint architecture, but here are the key highlights of each key technology component
- Global routing gateway that can enforce tenant data isolation at the session and connection level, route to the correct tenant, and provide a single database experience across all virtual tenant DBs
- Database clustering technology that helps place tenants on shared or dedicated Postgres compute anywhere globally and manage schemas migrations across tenants
- Tenant-aware Postgres pages that ensure pages are always isolated to a single tenant, which helps to move tenants instantaneously, generate backups per tenant/customer, fork the database with data of a specific customer for testing, and provide performance insights by tenants.
- Storage and compute decoupling to move tenant pages instantaneously across computes to provide instant capacity and manage performance guarantees for any tenants
- Cloud-native architecture by leveraging S3 for permanent storage and replication cross-region