Build on FoundryDB
FoundryDB is one EU-resident platform for the pieces most applications need: managed databases, managed GPU inference, object storage, and app hosting. They are designed to compose. This page maps the building blocks to real applications and points at the hands-on tutorial for each.
The building blocks
| Block | What it gives you |
|---|---|
| Managed databases | PostgreSQL, MySQL, MongoDB, Valkey, Kafka, OpenSearch. PostgreSQL includes pgvector for embeddings. |
| Managed inference | Open-weight LLMs (chat, embeddings, rerank) served on dedicated EU GPUs, behind an OpenAI-compatible endpoint. Fine-tuned LoRA adapters too. |
| Files | S3-compatible, EU-resident object storage for documents, model artifacts, and backups. |
| Stacks and app hosting | Launch a whole application and its backing services from one template, or deploy your own container. |
Two ways to start
- Get running fast with a Stack. One launch call provisions an app plus everything it needs, wired together. See Launch an App with Stacks.
- Build it yourself with the APIs. Provision each service and connect them in code, for full control over the architecture.
Common combinations
Each of these is a working application built from two or more blocks, with a tutorial that runs end to end.
Retrieval-Augmented Generation (RAG)
Managed PostgreSQL with pgvector holds your documents and their embeddings; managed inference serves both the embedding model and the chat model. Retrieval keeps answers grounded in your own data, and nothing leaves your EU resources.
→ Build an In-Platform RAG Assistant
A running app on your managed database
A Stack composes an application (a BI tool, a GraphQL API, a headless CMS, a no-code database UI) on top of a managed database, deployed on app hosting behind a managed TLS domain, in a single call.
Serve and fine-tune your own model
Deploy an open-weight LLM on a dedicated GPU, call it like OpenAI, then hot-load your own LoRA fine-tune with no restart and roll it back at will.
→ Deploy and Call Your First Model and Serve a Fine-Tuned LoRA Adapter
Bring your own model provider
Route to OpenAI, Anthropic, Mistral, Azure, or Groq through the same EU-enforced proxy and inference key, deciding managed-versus-BYO by the model prefix.
Data pipelines and analytics
Stream and transform data across engines, or auto-embed a table for semantic search.
→ Kafka to PostgreSQL Streaming, Build a RAG Pipeline, and Event Analytics
Where to go next
- Managed Inference for GPU plans, the curated catalog, and fine-tuning.
- AI Data Platform for vector search, embedding pipelines, and natural-language query.
- Deploy and Call Your First Model for the fastest path to a working inference endpoint.