Skip to content

Storage

SmartAGT separates structured data from files. PostgreSQL and pgvector hold the application data, the sessions and the indexes. Built-in storage or S3 holds the documents, attachments and generated files.

Structured data and files have different rolesStores are shared by the processes that need them
FrontendAccess to its
session database
BackendPersistence
APIConversations, configuration and filesSearch in the authorized indexes
WorkersDocuments, passages and embeddingsJobs and processing state
FilesBuilt-in: one server
S3: scalable
HTTPS
PostgreSQLsmartagt · smartagt_bffData and sessions
PostgreSQL + pgvectorsmartagt_vectorDocument indexes
Three logical databases can share one PostgreSQL instance

The Frontend service keeps its sessions in its database. The API and the relevant workers access the files over HTTPS.

Storage choice

File storage is chosen at installation, according to how the platform is expected to grow.

  • Built-in storage


    For a single server. SeaweedFS, included in the Backend package and managed by SmartAGT: no dependency or certificate to provide.

    • PoC and single-server installation
    • Files under the data directory of the Backend
    • Backup of this directory under your responsibility
    • Limited to a single server
  • S3


    Scalable. Your S3-compatible object service: MinIO, Ceph, NetApp StorageGRID or the S3 of a cloud provider.

    • Several servers and containers
    • High availability and backup carried by your storage
    • HTTPS endpoint, dedicated private bucket and access to provide
    • Connection before the installation: Configure S3

If the platform may one day extend to several servers, choose S3 from the installation.

PostgreSQL and pgvector

Database Use Main clients
smartagt Application data, configuration, identities, jobs and metadata of the internal storage API and workers
smartagt_vector Document indexes, vector extension API and RAG-Embed worker
smartagt_bff Frontend service sessions Frontend service

The reference profile uses PostgreSQL 18. The three databases can live in the same instance: their logical separation does not require three servers. pgvector adds vector search. It does not keep the original files.

Persistence and keys

Replacing the package or a container must find the databases, the files and the encryption keys again. The Backend and the Frontend service each keep their own secrets: see Secrets. Persistent data stays outside the binaries directory.

Generated images are copied to the managed storage: a provider URL is not a lasting retention. Retained images and OCR results follow the rights of their scope. Studio tests and conversation attachments have their own retention rules, described in the product.