← NovaCortex

Roadmap

What's shipped, what's next.

NovaCortex is the privacy-first, self-hostable, MCP-native, portable memory layer for AI agents. Your knowledge base never leaves your infrastructure, and a portable open format (PMF) means zero lock-in. Apache-2.0, free to self-host.

✓ Shipped

v1.0Self-host GA
  • ✓Graph memory with typed relations
  • ✓Semantic + full-text search
  • ✓MCP server
  • ✓REST API
  • ✓Web UI
  • ✓Docker / GHCR images
  • ✓CI pipeline
v1.0.1Stability
  • ✓Pagination
  • ✓Query filter
  • ✓PMF relation integrity
  • ✓Transaction-conflict retry
v1.1Portability + ops
  • ✓Binary (MessagePack) & AES-256-GCM-encrypted PMF
  • ✓Differential + streaming export
  • ✓Opt-in OpenTelemetry
  • ✓Recency-weighted search
  • ✓Vector reconcile
v1.2Open-core licensing
  • ✓Unforgeable ed25519-signed license keys (offline-validated)
  • ✓Public site
  • ✓Reproducible context-efficiency benchmark
  • ✓v1.3 Intelligence: LLM fact extraction with append-only supersession (fully-local capable)
  • ✓Hybrid retrieval (dense + BM25 RRF) with cross-encoder reranking
  • ✓LongMemEval_S: 80.0% (gpt-4o & gpt-5-mini readers), 63.4% fully offline — raw data published

→ Next

→

Multi-session aggregation + event-time filters

The two measured gaps from our LongMemEval error analysis: counting across many sessions and filtering by when things happened, not when they were stored.

→

Self-serve Pro

Stripe checkout that issues a signed license automatically — no manual steps.

→

Managed option, bring-your-own-infra

The convenience of a managed service without giving up the privacy guarantee — your data stays on your infrastructure.

→

Broader MCP tool coverage + SDK parity

More first-class MCP tools and consistent capabilities across every SDK.

This roadmap is directional, not a commitment. Priorities and timing may change. For real, day-to-day progress, follow the work on GitHub.

Star on GitHubBack to home