Topics Covered:

Enterprise Knowledge Engine (EKE): Intelligence layer orchestrating knowledge access
Vector Memory: Document and knowledge embedding system
Session Context: Temporary and persistent memory across interactions
Memory Items: Creation, modification, and management of knowledge base

Vector Memory

EverWorker uses Vector Memory to store and retrieve unstructured knowledge, such as documents, articles, or manual notes.

Supports semantic retrieval using vector embeddings
Each memory item is broken into chunks and embedded using models like nomic-embed-text or mxbai
Items include both an embedded body (used for semantic search) and a descriptive prompt (used in LLM context injection)
Memory entries can be uploaded files, scraped URLs, or manual entries
Supports metadata filters and token budgeting to control LLM context length

Session Context

Each interaction session maintains its own scoped Session Context - a mix of:

Enabled memory items
Uploaded content
Session settings and metadata

Users can actively manage session context via UI toggles, and Builders can configure which memory is injected automatically.

Memory Items

Memory Items are modular, reusable knowledge blocks. Key properties:

Can be edited, renamed, versioned
Include rich metadata (source, page count, tags, etc.)
Description fields shape how the item is referenced and injected into prompts
Access-controlled and assigned to workspaces or roles
Are the building blocks of Worker cognition

Summary

The memory system in EverWorker is designed to augment Workers with deep, contextual, and secure knowledge access. EKE drives smart orchestration. Vector Memory enables semantic understanding. Session Context keeps every interaction grounded and relevant. And as graph capabilities evolve, Workers will be able to reason across structured knowledge in more powerful, transparent ways.