Skip to content

Launch infrastructure

The rails behind
the vault.

Deskwise's launch setup uses AWS Amplify for the statically exported Next.js frontend, Render for the Django API and persistent data, and Amazon Bedrock for permission-filtered AI. Cloudflare manages DNS for the custom domains.

Launch topology · Amplify + Render + Bedrock

01

Web

AWS Amplify · static export

02

API + data

Render · Oregon

03

AI

Amazon Bedrock · India profile

04

DNS

Cloudflare · custom domains

The launch stack

One product.
Clear boundaries.

Cloudflare handles DNS, Amplify serves the web client, Render runs the stateful API, and Bedrock supplies managed AI. Each provider has a defined job.

01

Web delivery

Cloudflare manages DNS while Amplify builds and serves the static Next.js frontend through AWS edge delivery.

  • Cloudflare DNS
  • AWS Amplify Hosting
  • CloudFront CDN
  • HTTPS
02

API

Run Django and Gunicorn as a separate Render web service at `api.saadops.site`, with a public health endpoint and independent deploys.

  • Render Web Service
  • Django + Gunicorn
  • HTTPS custom domain
  • Health checks
03

Data

Store company records in Render PostgreSQL and keep uploaded documents/generated forms on a mounted persistent disk for the single-instance launch.

  • Render PostgreSQL
  • Persistent disk
  • Render backups
  • Restore checks
04

Intelligence

Send only permission-approved passages over TLS to Amazon Bedrock’s India inference profile, then return a cited answer.

  • Amazon Bedrock
  • Claude 3.5 Haiku
  • Titan Text Embeddings V2
05

Operations

Use provider deployment history, Render health/logs, database and disk recovery features, plus AWS budget alerts for Bedrock usage.

  • Amplify build history
  • Render logs
  • AWS Budgets
  • Amazon SES

This is the selected small-scale launch topology, not an AWS-only architecture. Render persistent disks are single-instance and prevent horizontal scaling; move documents to shared object storage before adding API replicas.

Resilience

Built for
the boring failures.

This is a lean launch topology, not a multi-AZ guarantee. The operating plan is to protect data, monitor the API, and move to shared object storage before adding replicas.

The frontend and API deploy separately.

Amplify can roll back the static frontend independently from Render. If the API is unavailable, public pages can still load, but authenticated product actions wait for the API.

A Render deploy needs care.

The launch API uses one instance with a persistent disk. Render disk-backed services do not provide zero-downtime deploys or horizontal scaling; schedule changes and keep a rollback/restore procedure.

Data needs to come back.

Use Render PostgreSQL backups and persistent-disk snapshots, and test a restore before inviting users. Keep a separate export of critical records and documents.

Security

The controls are
the architecture.

Read the product security model
01

Origins are explicit

The browser calls only the configured API origin. Django CORS and CSRF trust are limited to the Deskwise frontend domain.

02

Secrets stay server-side

Bedrock and SMTP credentials belong in Render secret environment variables—not in Amplify build variables or browser code.

03

Permissions before AI

Tenant and document permissions are checked before retrieval, so unauthorized passages are not sent to the model.

04

Operations stay visible

Review product audit history, Amplify build logs, Render API health/logs, and AWS Bedrock spend alerts.

Data boundary

Know where
the data sits.

The current launch is a hybrid deployment. The app database, files, and model calls have different providers and regions; the distinction should be explicit.

Launch data boundary

Hybrid
  • Static frontend delivery uses AWS Amplify/CloudFront; Cloudflare provides DNS for the custom domains.
  • Django API, PostgreSQL, and the document disk are hosted in Render’s Oregon (US West) region.
  • Bedrock chat uses the India inference profile; only permission-approved passages are sent over TLS.
  • This is not an all-data-in-India deployment: stored app data is in Oregon (US West) while model inference is in AWS India.

Organizations can keep using a local model or another supported cloud provider. For a strict India-only data-at-rest requirement, this Render-backed launch topology must be changed before making that claim.

Next

See the product these rails support.