System Architecture

BabelBird separates file collaboration from business records, storage, search and preview processing. Plan these services individually and add AI, OCR or integration services for the enabled modules.

Application and distributed storage architecture — Cluster illustration from the whitepaper. Node counts and redundancy depend on the implementation.

Service Layers

Layer	Responsibility	Deployment considerations
User interfaces	Web, desktop and mobile access, sync and mapped drives	Features vary by client and OS version
Application services	Identity, organizations, permissions, files, sharing, projects and approvals	Scale with workload and availability requirements
Database	Business records, permissions and file metadata	Required; replication differs between standard and HA deployments
File storage	File content and version data	Object storage, shared storage or mounted disks according to the design
Search and conversion	Indexes and previews	Capacity depends on file counts, formats and concurrency
Background processing	Queues and automated tasks	Monitor errors, retries and backlog
AI and OCR	Recognition, vector indexes, retrieval and model calls	Local inference or approved external model APIs

Client and Server Responsibilities

Clients handle local interactions, transfers, synchronization and mapped drives. Servers enforce permissions and persist cloud data. A cached file or local save is not proof of a successful upload; verify transfer status and cloud versions.

Technology stacks in the whitepaper describe the implementation at publication time. They do not guarantee identical dependencies across releases. Confirm compatibility and component licensing for the deployment.

Databases Are Still Required on Cloud Platforms

The database stores structured business and permission data; object storage stores file content. Deployment on Alibaba Cloud, Huawei Cloud or another platform still requires a database, on a server or an appropriate supported managed service. Object storage does not replace it.

Cache and monitoring components depend on scale and design. An HA or K8S example is not the mandatory component list for a basic installation. See Configuration advice.

Availability, Backup and Security

Replicas and redundant storage reduce single points of failure; independent backups support recovery after deletion, corruption or attacks. Test consistent restoration of databases, files, settings and certificates. Historical uptime is not a permanent zero-failure guarantee.

Use HTTPS and enforce permissions across search, preview, downloads and third-party calls. Define the data scope and security policy for sharing, external models and integrations.