AI Knowledge Workspace

Plateforme RAG full offline (Mac M1) : microservices Spring Boot + frontend Angular, LLM local via Docker Model Runner, Qdrant pour la recherche vectorielle.

En bref

• Java 21 / Spring Boot 3.3.x, Spring Cloud 2023.0.x, Spring AI (OpenAI starter)
• RAG: Model Runner (qwen2.5, mxbai-embed-large) + Qdrant
• Data: PostgreSQL, JPA + MapStruct
• Frontend: Angular 18 (BFF OIDC via Gateway)
• Orchestration: Docker Compose (dev / prod)

Note : Spring AI le plus recent supporte uniquement Spring Boot 3.3.x et Spring Cloud 2023.0.x.

Services

• frontend : Angular + Nginx, reverse-proxy /api, /oauth2, /login, /logout
• gateway : BFF Spring Cloud Gateway (OIDC, session, TokenRelay)
• keycloak + keycloak-certgen : IdP OIDC + generation de cert
• document-service : CRUD documents, extraction texte, PostgreSQL
• rag-service : ingestion, embeddings, Qdrant, chat
• user-service : profil utilisateur (/api/users/me)
• postgres, qdrant, modelrunner, toolbox

Prerequis (Mac M1)

• Docker Desktop (+ Model Runner active et TCP host-side)
• ~12GB RAM pour les modeles
• Modeles precharges: docker model pull qwen2.5 et docker model pull mxbai-embed-large
• JDK 21 + Maven 3.9 (si build hors Docker)
• Node 20 (si build frontend hors Docker)

Demarrage rapide

Option automatique

make ca-root
# Importer la CA locale: docs/ca-import.md
make all

Option pas a pas

make ca-rootvscode, comment effectuer un merge
# Importer la CA locale: docs/ca-import.md
make env
make hosts-keycloak
make pull-models
make run-models
make build up

Endpoints

# Frontend : http://localhost:4200
# Gateway : http://localhost:8080
# Keycloak : https://keycloak.local:8443
# Qdrant : http://localhost:6333
# Model Runner : http://localhost:12434/v1

Auth par defaut

• admin/admin123 (roles ADMIN,USER)
• user/user123 (role USER)

API (exemples)

• Docs: GET /api/documents, POST /api/documents (multipart file ou content)
• RAG: POST /api/rag/answer {query}, streaming SSE /api/rag/query

Build locaux (sans Docker)

cd backend && mvn clean package
cd frontend && npm install && npm run build

Docs utiles

• docs/ca-import.md : import de la CA locale
• docs/rag-call-flow.md : flux RAG (ingestion + query)