Fix: pre-create Neo4j dir structure, remove run dir redirect, lower memory

Fix: pre-create Neo4j dir structure, remove run dir redirect, lower memory

Fix: pre-create Neo4j dir structure, remove run dir redirect, lower memory

Fix: use sed to replace Neo4j conf keys, no duplicate declarations

Fix: no symlinks, chown /data/neo4j to neo4j user, direct console output to supervisord

Fix: install Neo4j+cypher-shell via official Debian apt repo (no tar.gz/APOC download)

Fix: Neo4j 5.24.0 + APOC 5.24.0 hardcoded (confirmed published)

Fix: Neo4j 5.26.0 (APOC available) + correct graphrag schema from seeder

Fix: install APOC plugin + allow apoc.* procedures in Neo4j config

Fix: always set Neo4j password on boot + cypher-shell fallback in seeder

Fix: backend waits for Neo4j bolt before starting uvicorn

Fix: upgrade Java 17→21 for Neo4j 2026.x compatibility

Fix: empty Neo4j startup + auto-seed on first boot

Fix: remove supervisord environment block so HF Spaces secrets reach uvicorn

Fix APOC: keep plugins in /opt/neo4j/plugins (not /data), use direct neo4j_conn for graph stats

Fix Neo4j auth: disable auth for demo deployment (NEO4J_AUTH=none), simplify entrypoint

Fix Neo4j auth: use NEO4J_AUTH env var (Neo4j 5.x standard), remove broken cypher-shell bootstrap

Fix 500: use 'or' fallback for env vars, guard eligible-patient query against Neo4j not ready

Fix backend crash: make Neo4jGraph lazy-init in graphrag.py — was failing at import time

Fix 502: retry Neo4j connection at startup, delay backend 50s to wait for Neo4j

Add Dockerfile to repo root for HuggingFace Spaces build

Initial deployment: ClinicalMatch AI v2.0 — FHIR R4 · MCP (9 tools) · A2A workflow · SHARP compliance · 100k synthetic patients · Neo4j graph · GraphRAG chatbot

initial commit