What is next on the CortexDB roadmap?

The roadmap for CortexDB targets retrieval accuracy gains to close the LoCoMo gap, temporal queries as first-class operations, 30+ data connectors by mid-year, and enterprise governance features like role-based access control and audit logging. CortexDB—the long-term memory layer for AI agents built by Apache Cassandra co-creator Prashant Malik—already ships a foundation of lossless event-sourced memory, 4-channel hybrid retrieval, and 93.8% on LongMemEval-S. The next phase expands the integration ecosystem and scales the higher-order memory operations.

Retrieval accuracy: closing the LoCoMo gap

CortexDB scores 93.8% on LongMemEval-S (469 of 500), ahead of Mem0's 93.4%. On LoCoMo (categories 1 to 4), Mem0 currently leads at 91.6% against CortexDB's 86.9%. Closing the 4.7-point LoCoMo gap is the focus of the next round of recall work.

Three workstreams are active.

Improved entity resolution. Better recognition of when different mentions refer to the same entity, especially across languages, informal references, and noisy source text. Asynchronous extraction will get a sharper resolver that improves the graph traversal channel in 4-channel hybrid retrieval.

Contextual ranking. Weighing recent memories differently from older ones when recency matters, while preserving long-term context when it does not. Cognitive Recall gets per-query-shape ranking signals.

Domain-specific tuning. Allowing teams to tune retrieval for their specific use cases (legal, healthcare, engineering, customer support) where domain vocabulary and relevance signals differ significantly. We are exploring overrides for specific domains during Cognitive Recall.

Temporal queries as first-class operations

Memory is not just what was stored. Memory is when. Temporal queries become first-class operations in the upcoming release.

• "What did we know about this customer last quarter?"
• "How has the team's sentiment about this feature changed over time?"
• "What was the state of this project on March 1st?"

Point-in-time recall reconstructs the full context as it existed at any moment in the past. We leverage our bi-temporal model (valid_from, valid_to, recorded_from, recorded_to) and the /v1/facts/timeline endpoint to explicitly support time-bounded recall paths, change-detection queries, and snapshot reconstruction. This matters for compliance, auditing, and tracing how decisions evolved across the corpus.

Expanded connector ecosystem

CortexDB ships 16 data connectors today. The roadmap targets 30+ by mid-year. Next on the integration queue:

• Postgres and MySQL for direct database ingestion as the source of truth for application state.
• Notion and Coda for product documentation and decision records.
• Calendly and Google Calendar for meeting context.
• Linear and Asana improvements with bidirectional sync.
• Custom webhooks for arbitrary event sources beyond the predefined connectors.

Every new connector inherits the same architectural properties: authentication and pagination handled, incremental sync, deduplication, and rich metadata that asynchronous extraction uses to build entity and relation edges.

Enterprise governance features

For teams deploying CortexDB in production at scale, the roadmap covers four enterprise concerns.

Role-based access control. Fine-grained permissions on who can read, write, and administer each partition. The hierarchical scopes in CortexDB already isolate memory at the storage layer; RBAC adds per-role policies on top of the scope boundary.

Audit logging. Complete trails of every memory access for compliance requirements. The lossless event-sourced log already records every write as an event; our audit endpoints (like /v1/audit) will record every read alongside the existing write log.

Data residency controls. Choose where data is stored to meet regulatory requirements across jurisdictions. CortexDB's storage layer becomes deployable by region.

SSO and SAML integration. Enterprise identity management for team deployments, plugging into the existing identity provider rather than the CortexDB account system.

Framework integration cadence

The AI framework ecosystem moves fast. We ship integrations for emerging frameworks and keep existing integrations current as frameworks release new versions. The current count of 53 integrations will grow on a regular cadence. Community requests directly influence the integration roadmap; the integrations currently scoped for the next release came from active production deployments asking for them.

Community-driven development

CortexDB is built for the developers who use it. The features above reflect feedback from early adopters, and the feedback loop stays tight.

• Discord for real-time discussions, feature requests, and direct access to the engineering team: discord.gg/cortexdb.
• Telegram for updates and conversation: t.me/cortexdb.
• GitHub Discussions for detailed feature proposals and technical conversations.

Several items on this roadmap started as community suggestions. The next batch will too.

What's next: V2 — the brain

V1 ships the four stages of the memory cycle that operate continuously: Capture, Index, Update, and Forget. V2 is about the higher-order operations—Consolidate and Compress. The "sleep" stages where the brain shows up. Where raw experience becomes understanding.

V2's targets: 96–97% on LongMemEval-S, 92–94% on LoCoMo, and p50 end-to-end answer latency under 4 seconds. The synthesizer is already in flight as a beta endpoint—call POST /v1/understanding/synthesize to trigger a consolidation pass over a scope today, and the concepts land in GET /v1/understanding.

Even the primitive version of the "sleep" stage is unique in the market. No competitor builds it. That's what makes CortexDB feel like a brain instead of a log—and the gap widens with every cycle.

Frequently asked questions

What is on CortexDB's next-release roadmap?

The next-release roadmap covers retrieval accuracy improvements targeting the LoCoMo gap, first-class temporal queries using our bi-temporal model, more data connectors (30+ by mid-year), and enterprise governance features (RBAC, audit logging, data residency, SSO and SAML).

When will CortexDB close the LoCoMo gap?

The 4.7-point LoCoMo gap (Mem0 91.6% against CortexDB's 86.9% on categories 1 to 4) is the focus of the next round of recall-side work. Three workstreams are active: improved entity resolution in asynchronous extraction, contextual ranking in Cognitive Recall, and domain-specific tuning.

What temporal query features are coming?

Point-in-time recall, change-detection queries, and snapshot reconstruction. We leverage our bi-temporal model and the /v1/facts/timeline endpoint to explicitly support time-bounded recall paths.

What new data connectors are coming?

The connector roadmap targets 30+ by mid-year. Next on the queue: Postgres and MySQL for direct database ingestion, additional documentation systems (Notion and Coda enhancements), calendar systems (Calendly and Google Calendar), and Linear and Asana improvements with bidirectional sync. Custom webhook connectors are also planned for arbitrary event sources.

What enterprise features are coming?

Role-based access control on top of hierarchical scopes, audit logging for reads, data-residency controls for regional storage requirements, and SSO and SAML integration for enterprise identity management.

How does CortexDB choose what to build next?

Community feedback directly influences the roadmap. Several roadmap items started as community suggestions on Discord, Telegram, or GitHub Discussions. The feedback loop stays tight because production deployments hit specific gaps and ask for specific features, and we build against the gaps that block production adoption.

Related reading