FastRAG — Hosted deployment (cloud profile) Diagram

Project configuration

Frontends: FASTRAG_API_URL + FASTRAG_QUERY_TOKEN, server-only (no NEXT_PUBLIC_); proxy route maxDuration 60s
API: FASTRAG_PROFILE=cloud, provider keys, DATABASE/REDIS/QDRANT URLs, FASTRAG_CALIBRATION_JSON pasted from the operator's calibration.json
Python bundle excludes web/, website/, tests/, docs/, eval/, scripts/ and *.md

Alternative: Render

render.yaml: one Docker web service, free plan, singapore, health check /health/ready
The Dockerfile boot script writes FASTRAG_CALIBRATION_JSON to disk and can download ONNX models first
Free plan spins down after ~15 minutes idle; no worker service in either hosted option

Limits

Demo and evaluation grade per docs/deployment.md: no HA, free-tier cold starts and LLM rate limits; cold starts rebuild the pipeline
Indexing never runs in the hosted API: it happens on the operator's machine