Glossary
Enterprise AI, defined plainly
The vocabulary of vector memory, agent orchestration, and self-hosted AI, without the hype.
- Vector memory engine
- A system that stores information as embeddings (numerical representations of meaning) so AI can recall relevant context semantically rather than by keyword. A vector memory engine adds persistence and governance: memory survives across sessions and access is scoped by permission.
- Embedding
- A list of numbers that represents the meaning of text, images, or other data. Two pieces of content with similar meaning have mathematically similar embeddings, which is what makes semantic search and memory recall possible.
- Agent orchestration
- Coordinating multiple specialized AI agents toward a shared goal: decomposing the task, assigning the right agent to each part, sequencing their work, and assembling the result. In VectorBrain this is the role of the Director engine.
- Persistent agent memory
- Memory that accumulates across sessions instead of resetting with each conversation. An agent with persistent memory knows what was decided last month, who owns which project, and how the organization talks. Context that compounds over time.
- RAG (Retrieval-Augmented Generation)
- A technique where an AI model retrieves relevant documents before generating an answer, grounding its response in real sources. RAG is retrieval at question-time; persistent agent memory goes further by accumulating and organizing knowledge continuously.
- Self-hosted AI
- AI infrastructure deployed inside an organization's own environment (VPC, datacenter, or air-gapped network) instead of a vendor's shared cloud. Prompts, data, and model traffic never leave infrastructure the organization controls.
- Air-gapped deployment
- A deployment with no connection to external networks, used in defense, classified, and critical-infrastructure settings. Requires locally hosted models and zero external API calls.
- Model routing
- Directing each AI task to the most appropriate model, by capability, cost, latency, or data-sensitivity policy. Model-agnostic routing prevents vendor lock-in and lets sensitive workloads stay on local models.
- Director agent
- In VectorBrain's architecture, the coordinating engine that translates a goal into a plan, dispatches specialist agents, and supervises execution: the conductor of the agent fleet.
- Memory scoping
- Partitioning AI memory by team, project, and role so agents can only recall what their invoking user is permitted to see. The memory-layer equivalent of access control.
- AI audit trail
- A complete, timestamped record of every agent action, memory access, and tool call: who triggered it, what it touched, and what it produced. The prerequisite for AI in any governed environment.
- Shadow AI
- Unsanctioned AI tool usage inside an organization, like employees pasting company data into consumer chatbots because no governed alternative exists. The risk self-hosted AI is deployed to eliminate.
Past the vocabulary stage?
See the concepts running as one engine, on your data, in your environment.