Key topics and definitions pertaining to AI and AgenticFort.

R = Retrieval
A = Augmented
G = Generation

RAG (Retrieval-Augmented Generation) is a technique that combines information retrieval with a generative AI model.

Instead of relying solely on what the model learned during training, RAG retrieves relevant documents or data from an external knowledge source at query time and feeds them into the model’s context, so its answer is grounded in that retrieved material.

The basic flow works in three steps:

1. Retrieve — the user’s question is used to search a knowledge base (often a vector database) for relevant passages, typically via semantic similarity.
2. Augment — those retrieved passages are inserted into the prompt alongside the original question.
3. Generate — the language model produces an answer conditioned on both the question and the retrieved context.

Why it matters:

• Reduces hallucinations — answers are anchored to real source documents rather than the model’s parametric memory.
• Fresh knowledge — the model can access up-to-date or proprietary information without retraining or fine-tuning.
• Citations/traceability — responses can reference the exact documents they drew from, which matters for enterprise and legal use cases.
• Cost efficiency — updating a knowledge index is far cheaper than retraining a model.

Common variants include hybrid search (combining keyword BM25 with vector similarity), reranking of retrieved chunks, and agentic RAG where the model decides when and what to retrieve.

Associated topics:

• Like embedding
• Chunking strategies
• Evaluating RAG quality

G = Governance 
R = Regulatory 
C = Compliance

AI is driving a key shift from backward-looking compliance checklists to forward-looking intelligence engines that provide real-time, enterprise-wide risk visibility.

The concept of “shadow AI” — employees using AI tools without official approval — has become a significant enterprise risk, mirroring the earlier “shadow IT” problem. This makes GRC oversight of AI not just a regulatory checkbox but a practical necessity for managing real exposure.

As one source (Continuum GRC) put it: 

“The next era of GRC will be defined by confidence and the ability to understand the risk posture and act before exposure materializes.”

AgenticFort delivers:
Actual LLM users chat monitoring and audit, models, organizations documents, prompts, answers, connections, decisions, artifacts and enforcement.

Vector Storage

A system purpose-built for storing, indexing, and searching high-dimensional numerical vectors — called embeddings.

In a RAG pipeline, it’s the component that makes step 1 (“Retrieve”) fast and semantically meaningful.

How it works:

1. Embedding generation — Documents are split into chunks, and each chunk is converted into a vector (e.g., 768 or 1,536 dimensions) by an embedding model. Text with similar meaning ends up close together in this vector space, regardless of exact wording.
2. Storage + indexing — Vectors (plus the original text and metadata) are stored in the database with a special index structure.
3. Similarity search — At query time, the user’s question is embedded the same way, and the store finds the nearest neighbors using a distance metric — typically cosine similarity, dot product, or Euclidean (L2) distance.

Chroma DB/Chroma

An open-source vector database designed to make embedding storage and similarity search simple, especially for AI application development.

It’s become a popular default choice for RAG prototypes and tutorials because it runs locally with almost zero setup.

Core characteristics:

• Embedded-first: By default it runs in-process within your Python or JavaScript app (pip install chromadb), persisting data to disk — no server to deploy. It can also run in client/server mode if you want to scale it out.
• Simple API: Creating a collection, adding documents, and querying takes just a few lines:

N = Named
E = Entity
R = Recognition

A subtask of Natural Language Processing that identifies and classifies specific items — called entities — in unstructured text into predefined categories.

It’s one of the foundational information-extraction tasks, and it underpins a lot of higher-level NLP and RAG work.

Artifacts

Some AI products use the term “artifacts” to describe rich outputs rendered alongside the chat. e.g. Claude Artifacts displays standalone code snippets, documents, or interactive React apps in a side panel rather than inline.

Conceptually, it’s about separating durable work products from ephemeral conversation.

O = Optical
C = Character
R = Recognition

The process of converting images of typed, printed, or handwritten text into machine-readable text.

It bridges the physical world (documents, photos, scans, PDFs) and the digital pipelines (RAG, NLP, search) we’ve been discussing.

Air-gapped

A system or network that is physically isolated from untrusted networks — most commonly the public internet.

The “gap” is literal: there is no wired or wireless connection between the isolated system and the outside world.