feat: Replace Weaviate knowledge base with unified memory system

.env.example

@@ -135,6 +135,17 @@ SEARXNG_SECRET=
 RCA_OPTIMIZE_COSTS=false
 GEMINI_DISABLE_THINKING=
 
+# Embedding model for alert correlation (optional).
+# Improves alert-to-incident grouping with semantic similarity.
+# Requires a DIRECT API key — OpenRouter does NOT support embeddings.
+# If left empty, uses Jaccard token similarity (works fine for most cases).
+# Format: provider/model-name
+# Examples:
+#   openai/text-embedding-3-small  (requires OPENAI_API_KEY)
+#   google/gemini-embedding-2      (requires GOOGLE_AI_API_KEY)
+#   bedrock/amazon.titan-embed-text-v2:0  (requires BEDROCK_ACCESS_KEY_ID)
+EMBEDDING_MODEL=
+
 # Multi-agent RCA orchestrator (opt-in). When false (default), background RCA
 # uses the legacy single-agent ReAct loop with RCA_MODEL. When true, a lead
 # orchestrator triages each RCA and may fan out parallel read-only sub-agents.
@@ -333,12 +344,6 @@ USE_UNTRUSTED_NODES=
 NEXT_PUBLIC_KUBECTL_AGENT_CHART_URL=
 
 # -----------------------------------------------------------------------------
-# Weaviate (Vector Database)
-# -----------------------------------------------------------------------------
-WEAVIATE_HOST=weaviate
-WEAVIATE_PORT=8080
-WEAVIATE_GRPC_PORT=50051
-
 # -----------------------------------------------------------------------------
 # SSL/TLS for External Services
 # -----------------------------------------------------------------------------
@@ -351,9 +356,6 @@ POSTGRES_SSLROOTCERT=
 REDIS_SSL_CA_CERTS=
 REDIS_SSL_CERT_REQS=
 
-# Weaviate: set to true for external instances with HTTPS+gRPC-TLS
-WEAVIATE_SECURE=false
-
 # -----------------------------------------------------------------------------
 # Memgraph (Graph Database)
 # -----------------------------------------------------------------------------

.github/workflows/publish-airtight.yml

@@ -52,8 +52,6 @@ jobs:
         amazon/aws-cli:2.34.6
         memgraph/memgraph-mage:3.8.1
         memgraph/lab:3.8.0
-        cr.weaviate.io/semitechnologies/weaviate:1.27.6
-        cr.weaviate.io/semitechnologies/transformers-inference:sentence-transformers-all-MiniLM-L6-v2
 
     steps:
       - name: Free disk space

AGENTS.md

@@ -17,10 +17,10 @@
 - **Important**: Always update both `docker-compose.yaml` and `docker-compose.prod-local.yml` together to keep environment variables in sync.
 
 ## Architecture
-- **Docker Compose stack**: aurora-server (Flask API on :5080), celery_worker (background tasks), chatbot (WebSocket on :5006), frontend (Next.js on :3000), postgres (:5432), weaviate (vector DB :8080), redis (:6379), vault (secrets :8200), seaweedfs (object storage :8333)
+- **Docker Compose stack**: aurora-server (Flask API on :5080), celery_worker (background tasks), chatbot (WebSocket on :5006), frontend (Next.js on :3000), postgres (:5432), redis (:6379), vault (secrets :8200), seaweedfs (object storage :8333)
 - **Backend** (server/): Flask REST API (main_compute.py), WebSocket chatbot (main_chatbot.py), Celery tasks, connectors for GCP/AWS/Azure/Datadog/New Relic/Grafana, LangGraph agent workflow
 - **Frontend** (client/): Next.js 15, TypeScript, Tailwind CSS, shadcn/ui components, Auth.js authentication, path alias `@/*` → `./src/*`
-- **Database**: PostgreSQL (aurora_db), Weaviate for semantic search, Redis for Celery queue
+- **Database**: PostgreSQL (aurora_db), Redis for Celery queue
 - **Secrets**: HashiCorp Vault (KV v2 engine at `aurora` mount)
 - **Object Storage**: S3-compatible via SeaweedFS (default), supports AWS S3, Cloudflare R2, MinIO, etc.
 - **Config**: Environment in `./.env`, GCP service account in `server/connectors/gcp_connector/*.json`

README.md

@@ -271,7 +271,6 @@ aurora/
 | Backend | Python, Flask, Celery |
 | Frontend | Next.js, TypeScript |
 | Graph DB | Memgraph |
-| Vector Store | Weaviate |
 | Secrets | HashiCorp Vault, AWS Secrets Manager |
 | Storage | PostgreSQL, Redis, S3-compatible |
 

client/next.config.ts

@@ -23,6 +23,12 @@ const nextConfig: NextConfig = {
   eslint: {
     ignoreDuringBuilds: true,
   },
+  experimental: {
+    serverActions: {
+      bodySizeLimit: "50mb",
+    },
+    middlewareClientMaxBodySize: "50mb",
+  },
   cacheMaxMemorySize: 0,
   poweredByHeader: false,
   webpack: (config, { isServer, dev }) => {

client/src/app/api/proxy/knowledge-base/[...path]/route.ts → client/src/app/api/proxy/memory/[...path]/route.ts

@@ -6,8 +6,8 @@ async function handler(
   { params }: { params: Promise<{ path: string[] }> },
 ) {
   const { path } = await params;
-  const backendPath = '/api/knowledge-base/' + path.join('/');
-  return forwardRequest(request, request.method, backendPath, 'knowledge-base');
+  const backendPath = '/api/memory/' + path.join('/');
+  return forwardRequest(request, request.method, backendPath, 'memory');
 }
 
 export { handler as GET, handler as POST, handler as PUT, handler as DELETE };