Cortex
Organizational Memory Platform for AI Agents
Cortex
Problem
Enterprise knowledge is fragmented across Slack, GitHub, Jira, and Linear. Agents that only see one silo invent context, lose provenance, and cannot be trusted in production workflows.
Approach
Built an MCP-based organizational memory platform that ingests collaboration events through Kafka and serves structured context to AI agents through FastAPI and Neo4j.
Implemented hybrid retrieval with Neo4j full-text search and Qdrant cosine similarity over normalized sentence-transformer embeddings.
Added Redis caching, provenance controls, RBAC, GDPR erasure, Prometheus metrics, and OpenTelemetry tracing for reliable production retrieval.
Results
MCP memory · Kafka ingest · hybrid retrieval
- Reduced fragmented retrieval across enterprise knowledge sources for agent tool-use
- Hybrid graph + vector retrieval with provenance-aware responses
- Production controls: caching, RBAC, erasure, metrics, and distributed tracing
Stack
What I learned
Agent memory is an infrastructure problem before it is a prompting problem. Provenance, access control, and retrieval latency decide whether teams trust what the agent recalls.