Browse the documentation
AI-Soda Brain

Preparing and validating a private deployment

Map the data and identity boundaries before installing, then validate the full path from the user interface to the data source. After startup, test permissions, queries, citations and recovery to establish that the system is usable for business work.

Technical guide · November 20263 min read

Define the deployment boundary

List the locations of application services, model inference, business data, document indexes, logs and backups. For every connection across a network boundary, record its direction, purpose, authentication method and owner. If processing must stay inside a designated zone, include inference, diagnostic logs, remote support and backups in the review.

Size the environment using document volumes, query complexity, concurrent active users and response targets. A total user count or the hardware used for a single demonstration is insufficient to choose a configuration for every project.

Preinstallation checks

CheckWhat to confirm
Version and distributionApproved package source, version and integrity verification
Environment and dependenciesSupported operating systems, runtimes, models and storage dependencies
Network and certificatesDomain names, certificates, clock synchronization and required connectivity
Data accessA dedicated read-only account with access to the minimum necessary tables and fields
Operational readinessBackup location, recovery ownership, change window and rollback plan

Validate the full path

  1. Follow the delivery manual for the selected release and retain the configuration record.
  2. Verify service health and sign-in, then access a small test dataset with the read-only identity.
  3. Ask a question with a reference answer and check the source, period and units.
  4. Switch to a test identity with fewer permissions. Verify its result scope and how the system handles requests it must decline.
  5. Perform a controlled restart and a backup restoration drill. Confirm that configuration and indexes remain usable.

Diagnose by symptom

If the entry point is unreachable, check the network, domain, certificate and service status. If sign-in works but data is missing, check identity mapping, source permissions and synchronization. If a query runs but the answer differs, check the snapshot and metric definition. Diagnose the cause before changing account permissions.

This guide covers preparation and acceptance. Use the official documentation for the delivered release for installation commands, ports, images, minimum requirements and the support matrix.

Try “permissions,” “data source,” “manufacturing” or “API.”