Persistent AI Memory
Let agents recall relevant history across sessions instead of starting over.
Interactive platform map
Architecture in context
The focused blueprint, its required foundation, and declared recommendations.
Persistent AI Memory
Let agents recall relevant history across sessions instead of starting over.
What this replaces
Even powerful agents lose continuity when a session ends, forcing people to repeat project history, decisions and context.
What your team gains
A semantic session-history layer that can be searched through MCP or HTTP and injected into active work.
What is inside the blueprint
How it fits the platform
Conversation/session history -> embeddings -> recall.chunks in pgvector -> session-recall-mcp -> AI clients/personas.
session-recall-mcp, PostgreSQL, pgvector, embedding endpoint
- AICORTEX Core PlatformSession history and embeddings persist in Core's data tier.
From prerequisites to operation
- Captured session history
- Embedding pipeline
- Memory access/privacy policy
- Create recall schema/index
- Ingest historical sessions
- Deploy MCP/HTTP service
- Register with supported clients
- Create memory retrieval instructions
- Chunking
- Namespaces/tenants
- Similarity threshold
- Maximum recall context
- Retention and deletion rules
- Authentication
- Index growth
- Privacy boundaries
- Duplicate/stale memories
- Embedding migrations
- Retention enforcement
What this unlocks with other layers
Institutional Memory
The AI can retrieve documents, recall prior work and trace relationships and provenance.
AI SRE Foundation
Agents can reason over the same telemetry and history humans use for diagnosis.