Skip to content

RNR Enhanced Cognee

Enterprise-grade AI memory infrastructure with multi-tenant support, 30 pluggable storage providers across 4 tiers, 122 MCP tools, and 4 cross-language client SDKs.

License Python Tests Coverage Providers SDKs

What is RNR Enhanced Cognee

RNR Enhanced Cognee is a fork of Cognee that adds production-grade infrastructure: multi-tenant data partitioning, 30 pluggable storage providers, four official client SDKs, MCP server hardening (API-key + rate limit + payload cap), HTTPS/TLS support, encryption at rest, GDPR tooling, backup + recovery, and SigNoz + Apache Superset observability.

It exposes 122 MCP tools (Model Context Protocol) for AI IDEs like Claude Code, Cursor, and Continue, plus a FastAPI HTTP variant for cross-language SDK use.

Default stack -- 100% MIT + Apache-2.0

Tier Default provider License
Graph ArcadeDB Apache-2.0
Vector Qdrant Apache-2.0
Cache Valkey Apache-2.0
Relational PostgreSQL + pgvector PostgreSQL License

Every tier accepts a drop-in alternate via ENHANCED_*_PROVIDER env vars -- see Pluggable DB Backends for the full matrix.

Key features

Memory + Knowledge Graph

  • 122 MCP tools across 25+ categories (CRUD, search, dedup, summarisation, GDPR, backup, undo/redo, ...)
  • Semantic search with vector embeddings (Qdrant + pgvector + LanceDB + Chroma + Weaviate + Milvus)
  • Knowledge graph ingestion (cognify) with 9 graph backends including ArcadeDB / Apache AGE / Memgraph / Kuzu
  • Memory deduplication, summarisation, consolidation, tiering, provenance, confidence scoring, importance scoring
  • Versioning with undo / redo
  • Real Redis-backed personalization (agent affinity + interaction recency + query affinity)

Production Infrastructure

  • Multi-tenant data partitioning (TenantContext + per-tenant Postgres schema + HTTP X-Tenant-ID middleware)
  • Cross-language client SDKs: Python on PyPI; Node / Go / Rust registry-ready
  • MCP server hardening: API-key auth + per-tool token-bucket rate limit + payload cap
  • HTTPS/TLS support on the FastAPI server
  • Encryption at rest (Fernet AES-128-CBC) with key rotation
  • Audit logging + GDPR (consent, export, delete, tenant-isolation verifier)
  • Backup + verification + recovery + transaction manager
  • 4,661 tests passing, 95% coverage

Observability

  • SigNoz APM (replaces Grafana / Loki / Tempo / Jaeger -- all Apache-2.0)
  • Apache Superset BI with 5 importable dashboards
  • Prometheus metrics, OpenTelemetry tracing
  • LLM cost tracker with per-model budgets

Performance

Real Locust benchmark against the live default stack (ArcadeDB + Qdrant + Valkey + PostgreSQL):

Metric Value
RPS 48.86
p50 latency 11 ms
p95 latency 14 ms
p99 latency 17 ms
Error rate 0.00%

60s run with 20 concurrent users + 5/sec spawn rate. See tests/benchmarks/baselines/.

Get started

git clone https://github.com/vincentspereira/RNR-Enhanced-Cognee.git
cd RNR-Enhanced-Cognee
docker compose -p RNR-Enhanced-Cognee -f config/docker/docker-compose-enhanced-cognee.yml up -d
python -m uvicorn src.enhanced_cognee_mcp:app --port 8080

Then point Claude Code's MCP config at bin/enhanced_cognee_mcp_server.py. Full setup in Quick Start.

License

Apache-2.0.