#
PHOENIQS Data Pools
Retrieval-augmented generation (RAG) lets your AI assistants and automated workflows answer questions from your own documents and data, not just from the model's training knowledge. PHOENIQS Data Pools is a secure, scalable Swiss-sovereign RAG and retrieval workspace that connects your documents and structured data to AI chat, agents, and automated workflows. It is the retrieval layer behind PHOENIQS Chat, built for document handling, controlled access, and automation.
Data Pools sends content to the configured AI endpoint for embeddings, descriptions, reranking, and OCR. On PHOENIQS-hosted Chat instances that endpoint is the PHOENIQS AI Model Service on Swiss infrastructure. In other deployments, data-residency and model-training guarantees depend on the configured AI provider.
#
Key capabilities
#
Security and access control
User-content tables are protected at the database level using PostgreSQL Row-Level Security (RLS), scoped to the authenticated user's Keycloak identity. Access is enforced by the database itself, not only by the application layer.
Users can access pools they own, pools explicitly shared with them, and pools made public to authenticated users.
Stored API keys are encrypted at rest. Deleting a pool or a document uses cascading hard deletes for all related database records.
#
Retrieval quality
Data Pools runs hybrid search, combining semantic vector search with full-text retrieval using Reciprocal Rank Fusion. Chunking strategies, embedding models, retrieval settings, reranking, and OCR models are configurable per pool, so retrieval can be optimised for the type of content you are working with.
#
Document handling
- Unstructured documents: PDFs, Office documents, text files, HTML, and images. Scanned PDFs and images are processed by a configured vision-model OCR service, with Tesseract as a fallback.
- Structured data: CSV and TSV files are parsed into typed PostgreSQL-backed tables with a filtering DSL, sorting, pagination, and aggregations.
#
Built for agents and automation
Data Pools is not tied to the chat interface. It exposes a read-only MCP server with nine tools, session-authenticated REST endpoints, and a web UI, so AI agents and automated workflows can query shared knowledge pools directly.
#
Storage
Data Pools enforces a configurable per-user storage quota across uploaded documents and imported CSV and TSV tables. The default quota is 5 GB per user and can be changed by the deployment administrator.
Infrastructure storage capacity, commercial plan limits, and options for adding storage are deployment-specific. For the storage included with your PHOENIQS Chat plan, see Subscriptions.
#
How it works
Data Pools supports two types of knowledge pools.
RAG Pools are for unstructured documents such as PDFs, Office documents, text, HTML, and images. They use pgvector with hybrid search, vision-model OCR with Tesseract fallback, and configurable chunking and embedding models per pool.
Database Pools are for structured CSV and TSV files. Files are parsed into typed PostgreSQL-backed tables with a filtering DSL (eq, neq, gt, gte, lt, lte, contains), sorting, pagination, and aggregations.
An AI-maintained catalog generates pool and content descriptions and extracts structured facets such as people, organisations, topics, locations, and timeframes. Catalog generation is best-effort and does not block uploads if the configured AI endpoint is unavailable.
#
Access surfaces
- MCP server: nine read-only tools for agents and automation.
- REST API: session-authenticated endpoints used by the web application for pool and content management.
- Web UI: management interface for creating, configuring, sharing, and managing pools.
#
Getting started
- Provisioning. Storage capacity and availability depend on the deployment's configuration.
- Create a pool. In the web UI, create a RAG Pool for documents or a Database Pool for CSV or TSV data.
- Add data. Upload documents or structured files to the appropriate pool.
- Connect your agents. Point your agent or workflow at the MCP endpoint to discover and query the pools available to the authenticated user.
- Manage storage. Monitor the per-user quota in the web UI. Additional infrastructure storage is managed according to the deployment's commercial and infrastructure configuration.