Apcone - Serviço RAG Minimalista
Adedev-W/Apcone
Serviço RAG minimalista com FastAPI, PostgreSQL, Redis e Qdrant
Ver no GitHubMCP Server que implementa um serviço FastAPI completo para Retrieval-Augmented Generation (RAG) com PostgreSQL, Redis e Qdrant. Oferece ingestão, busca e indexação de documentos com suporte a múltiplos tenants e OCR de PDFs via gRPC.
Instalação
npx convez-ai Adedev-W/Apcone1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Construir agentes IA com conhecimento contextual de documentos
- Implementar busca semântica em bases de conhecimento multi-tenant
- Processar e indexar PDFs escaneados com OCR automático
- Criar pipelines de ingestão de documentos com reindexação sob demanda
Veredito

Solução robusta e bem-estruturada para RAG em produção. Combina componentes maduros (Qdrant, PostgreSQL, Redis) com arquitetura clara de workers e suporte a tenants isolados. Ideal para quem precisa de infraestrutura RAG self-hosted escalável.
README
A tradução para o português deste README ainda está na fila.
apcone
Minimal FastAPI service with PostgreSQL, Redis, and Qdrant.
Setup
uv sync
docker compose up -d
uv run alembic upgrade head
uv run uvicorn app.main:app --reload
The default host ports are 5433 for PostgreSQL, 6380 for Redis, 6333 for Qdrant HTTP/UI, and 6334 for Qdrant gRPC.
RAG defaults:
QDRANT_COLLECTION=rag_chunksEMBEDDING_MODEL=BAAI/bge-small-en-v1.5CHUNK_SIZE=1000CHUNK_OVERLAP=150SEARCH_TOP_K=5
Health checks:
curl http://127.0.0.1:8000/health
curl http://127.0.0.1:8000/health/postgres
curl http://127.0.0.1:8000/health/redis
Qdrant checks:
curl http://127.0.0.1:6333/healthz
curl http://127.0.0.1:6333
Qdrant dashboard is available at http://127.0.0.1:6333/dashboard.
Core endpoints:
POST /documents/ingestPOST /documents/uploadPOST /documents/upload-documentPOST /documents/searchGET /documents/jobs/{job_id}GET /documents/{document_id}GET /documents/{document_id}/chunksPOST /documents/{document_id}/reindexDELETE /documents/{document_id}
Document APIs and MCP tools support tenant-scoped knowledge namespaces with
tenant_id and scope. Existing calls default to tenant_id=default and
scope=default; pass explicit values when multiple teams, projects, or agent
workspaces share the same Apcone service.
Worker:
uv run python -m app.workers.run_worker
The worker uses a short RQ heartbeat/dequeue cycle by default
(RQ_WORKER_TTL_SECONDS=120) so an idle queue does not leave a Redis blocking
read open long enough to be closed by the connection path. Redis client
timeouts can be tuned with REDIS_SOCKET_TIMEOUT_SECONDS,
REDIS_SOCKET_CONNECT_TIMEOUT_SECONDS, and
REDIS_HEALTH_CHECK_INTERVAL_SECONDS.
PDF OCR fallback runs through the Alpine-based gRPC scanner service:
docker compose up -d pdf-scanner
The worker calls PDF_SCANNER_GRPC_URL, which defaults to 127.0.0.1:50051.
PDF_RESULT_TTL_SECONDS only controls how long Redis keeps a completed job result; it is not scan duration.
PDF ingestion profiles each document before choosing a parser:
- simple text PDF: PyMuPDF
- table/layout candidate: pdfplumber for the candidate pages
- scanned or mixed PDF: OCRmyPDF only for detected scanned pages
For high traffic, run more app.workers.run_worker processes for regular parsing, but keep OCR capacity near available CPU with PDF_SCANNER_WORKERS and PDF_SCANNER_OCR_JOBS. Text-only OCR is the default path for RAG indexing; searchable OCR PDF generation should be treated as a separate heavy operation.
If PDF_SCANNER_USE_SOURCE_PATH=true, mount the upload storage into the scanner container and set PDF_SCANNER_ALLOWED_DIR to that mounted directory.