A structured, shared store of an organisation's facts, terminology, processes and relationships that AI agents read from and write to.
Unlike a document repository, it represents knowledge as governed entities rather than unstructured text, so multiple agents can act on a consistent view of the business.
See also: Company knowledge layer for AI agents
The condition where multiple AI agents in the same organisation operate from different sources of truth for shared concepts, such as pricing, policies, customer state or terminology, and produce confidently different answers to the same question.
Each agent faithfully represents the data it was given; the problem is that nobody gave them the same data.
See also: Get your agents on the same page
A memory layer that every agent in an organisation can read from and contribute to, rather than per-agent silos.
When one agent learns a new fact, every other agent has access to it immediately, scoped by permissions.
See also: Shared agent memory
Memory stored as typed entities and explicit relationships rather than as unstructured text chunks.
Structured memory supports precise queries, governance and versioning, and avoids the ambiguity that comes from retrieving free-form text.
See also: Oiya vs Zep and Graphiti
A software system in which two or more AI agents, often built on different frameworks or models, operate within the same organisation and need to coordinate on shared context.
Multi-agent systems introduce alignment problems that single-agent architectures do not have.
See also: How Oiya works
The layer that decides which agent runs, in what order, and with which inputs.
Orchestration governs control flow between agents but does not by itself guarantee that those agents share the same underlying facts or definitions.
See also: Oiya integrations
A pattern where an LLM is given relevant text chunks retrieved from a vector store at query time and uses them to generate an answer.
RAG improves relevance for a single query but does not enforce consistency across multiple agents or model the relationships between concepts.
See also: Oiya vs RAG
A data structure that represents entities as nodes and relationships as edges.
Traditional knowledge graphs are manually built and maintained, which causes them to decay as the business changes. A company knowledge layer adds continuous learning and governance on top.
See also: Oiya vs knowledge graphs
An abstraction that defines the meaning of business concepts, such as 'qualified lead' or 'active customer', independently of any single data source or query.
AI agents that share a semantic layer give consistent answers about the same concept across systems.
See also: Data fabric for AI agents
Knowledge stored outside any single LLM provider, accessible via API, so agents can switch between models (OpenAI, Anthropic, Google, open-source) without losing organisational context.
The opposite of provider-locked memory.
See also: Oiya vs OpenAI memory
The maximum amount of text an LLM can consider in a single inference.
Context windows are finite, which is why agents must rely on an external knowledge layer rather than holding all relevant business context in-prompt.
See also: Knowledge management for AI
The practice of supplying an LLM with verified, source-linked context at inference time so its output can be traced back to authoritative facts.
A structured knowledge layer improves grounding by providing typed entities and provenance instead of raw text chunks.
See also: Oiya vs Pinecone Nexus
The gradual change in a single agent's behaviour over time as models update, prompts evolve and underlying data shifts.
Agent drift is well-studied at the per-agent level; cross-agent divergence is a separate, less-observed problem.
See also: Oiya vs per-agent memory
The controls that determine which facts enter shared agent memory, who approves them, and how changes are audited and rolled back.
Without governance, every agent interaction risks polluting the knowledge layer with unverified information.
See also: Security and deployment