REFERENCE

Glossary

This glossary defines the concepts behind Synq's AI context layer: general terms like knowledge graph and retrieval-augmented generation, linked to their Wikidata entries, alongside terms specific to how Synq builds this layer, like context drift and institutional knowledge loss, defined in the language Synq actually uses.

Institutional knowledgeCaptured before it leaves.

Knowledge graphEntities and relationships.

Permissioned retrievalFetched under access rules.

Grounded answerAnchored in that context.

ProvenanceTraced back to the source.

Provenance leads back to source.

Agentic AI

Agentic AI is artificial intelligence that executes multi-step work and takes actions inside an organization's systems, rather than only answering questions or generating text for a human to act on.

Synq's context layer grounds agentic execution in an enterprise's real data and workflows, so agents act on what the organization actually knows rather than generic training data. Every action stays logged and reviewable, which is what makes autonomous execution auditable rather than a black box.

#agentic-ai

AI context layer

The AI context layer is the tier between an enterprise's raw systems and its AI agents that turns scattered institutional knowledge into governed, structured, provenance-tracked context a model can act on.

Synq builds this layer to keep AI grounded in how a business actually works as systems change and knowledge decays. It differs from plain retrieval-augmented generation by adding an enterprise knowledge graph, provenance, freshness, and permissions, so every answer is traceable and governed rather than just retrieved.

#ai-context-layer

Audit trail

An audit trail is a chronological record of a sequence of activities, kept so that any query, approval, or access can be reconstructed and reviewed after the fact.

On Synq, every query, approval, and access is logged and traceable end to end. The audit trail is what turns provenance from a claim into something a compliance officer can actually check after the fact.

#audit-trail

Context drift

Context drift is what happens when an AI context layer goes stale because the underlying enterprise systems and facts it was built from keep changing.

Synq treats context drift as an open research problem: detecting when grounding has gone out of date and correcting it before an AI system starts reasoning from outdated context.

#context-drift

Data governance

Data governance is the set of rules, roles, and processes an organization uses to ensure the quality, security, and proper use of its data, including who may access it and how.

In an AI context layer, governance determines who can see which data and how an AI agent's access maps to those same rules, so automation never bypasses the controls a human user would face.

#data-governance

Data lineage

Data lineage is the record of where a piece of data originated and every transformation or movement it went through before reaching its current form.

Lineage is closely related to provenance: where provenance traces an AI answer back to its source, lineage traces the data itself through the systems and pipelines it passed through to get there.

#data-lineage

Data sovereignty

Data sovereignty is the principle that an organization retains decision-making authority over its own data, including where that data physically resides and which jurisdiction's laws govern it.

Synq supports data isolation and residency, keeping retrieval and data inside your own environment, in-country where required. Your enterprise context is never pooled with anyone else's and never used to train a shared model.

#data-sovereignty

Durable agent memory

Durable agent memory is shared context that persists across sessions, so a multi-agent system does not relearn the same information on every run before it can do useful work.

Synq lists multi-agent memory as an open research problem: shared, durable memory so agentic AI does not relearn the same context each run. The related benchmark tests whether multi-agent systems retain shared context across sessions.

#durable-agent-memory

Enterprise knowledge graph

An enterprise knowledge graph is a structured model of an organization's entities, relationships, and decisions, built from its real systems and documents so an AI can follow explicit connections instead of guessing.

This is the graph stage of Synq's context-layer pipeline: capture, structure, graph, retrieve, ground, and serve. It is what lets a model reason about how a customer, a contract, and a workflow actually relate to each other inside a specific business.

#enterprise-knowledge-graph

Grounding

Grounding is the practice of anchoring an AI model's answers in real, retrieved context and citations, rather than relying only on the model's fluency alone.

A fluent answer and a grounded one are not the same thing: fluency measures how convincing the text sounds, while groundedness measures whether it is actually supported by real, traceable source material. Synq treats measuring true groundedness as an open research problem.

#grounding

Human approval gate

A human approval gate is a checkpoint that holds a sensitive action until a person reviews and approves it, so an agent cannot execute consequential work unsupervised.

On Synq, sensitive actions route through staged human review before they execute, passing L1, L2, and L3 review before anything runs. Every access, approval, and change made through those gates is logged for audit.

#human-approval-gate

Institutional knowledge loss

Institutional knowledge loss is the silent disappearance of an organization's operational knowledge as people leave, systems change, and the provenance behind past decisions is never captured.

Synq frames this as the failure mode nobody names until it is too late, and positions the AI context layer as the mechanism that captures institutional knowledge before it disappears rather than after.

#institutional-knowledge-loss

Intelligent agent

An intelligent agent is a system that perceives its environment and takes autonomous actions to achieve a goal, and that may improve its performance over time through learning or acquired knowledge.

This is the general computer-science definition behind what Synq calls agentic AI: the broader research concept an enterprise deployment specializes for a specific organization's context and guardrails.

#intelligent-agent

Knowledge graph

A knowledge graph is an information repository structured as a network of entities and the relationships that connect them, rather than as flat tables or documents.

Knowledge graphs let a system query relationships directly, such as which customer is linked to which contract and which decision, instead of inferring them from unstructured text. Synq's enterprise knowledge graph is a specific, enterprise-scoped application of this general concept.

#knowledge-graph

Large language model

A large language model, or LLM, is a language model trained on very large amounts of text, giving it broad ability to generate and reason about language.

LLMs are the reasoning engines behind most modern AI agents. Synq's context layer is model-agnostic: it grounds whichever LLM an enterprise chooses in that organization's real data, rather than being built around one specific model.

#large-language-model

Ontology

In information science, an ontology is a formal specification of a conceptualization: a defined set of concepts, categories, and relationships within a domain that both people and machines can share.

Ontologies give a knowledge graph its structure by defining what kinds of entities and relationships are allowed to exist in it. This is the information-science sense of the word, distinct from ontology as a branch of philosophy.

#ontology

Permissioned retrieval

Permissioned retrieval means an AI agent can only retrieve the context the requesting user is already entitled to see, so automation never widens access beyond a person's own permissions.

Access on Synq is role-based and permissioned, so an agent only retrieves what the requesting user is entitled to see. Permissions scale by role, from read-only to full administration, and every retrieval is logged.

#permissioned-retrieval

Provenance

Provenance is the traceable path from a grounded AI answer back to the original source document it was drawn from, so any claim can be checked against where it came from.

Synq treats provenance as a requirement rather than a feature: every answer in the context layer carries a citation from source to output, which is what makes an audit trail possible.

#provenance

Retrieval-augmented generation

Retrieval-augmented generation, or RAG, is a technique that lets a language model retrieve relevant information from an external source and incorporate it into its answer, rather than relying only on what it learned during training.

Synq's context layer is often compared to RAG because both retrieve external information, but a context layer adds an enterprise knowledge graph, provenance, freshness, and permissions on top of retrieval, so results follow real relationships rather than just text similarity.

#retrieval-augmented-generation

Semantic layer

A semantic layer is a business representation of data that maps raw technical fields to the terms and meaning people in an organization actually use.

A semantic layer is what lets a term like revenue or customer mean the same specific thing everywhere an AI agent or a person queries it, instead of every system defining those terms slightly differently.

#semantic-layer

Sovereign AI

Sovereign AI is an AI architecture that keeps context, data, and retrieval entirely inside an organization's own boundary, rather than routing through third-party infrastructure at all.

Synq treats sovereignty as both a technical open problem and a governance promise: control over your data, visibility into every decision, and the option to run entirely within your own environment.

#sovereign-ai