Privata Docs

Appearance

Glossary ​

This glossary explains common AI terms used throughout the documentation as well as the main platform terms used in Privata.

INFO

Some terms in this glossary describe general AI industry concepts. Those definitions are included to help users understand the ideas behind private AI, retrieval workflows, and document-based response generation.

General AI Terms ​

AI Model ​

An AI model is the software system that generates text, images, summaries, and other outputs based on the instructions and context it receives.

Completion ​

A completion is the output returned by an AI model after it receives a prompt. In practice, a completion can be an answer, a summary, a proposal draft, an image, or another generated result.

Context ​

Context is the information provided to the AI system along with the prompt. This can include user instructions, retrieved document content, source document text, or project-specific guidance.

Context Window ​

The context window is the amount of text or information a model can consider in a single request. If too much information is supplied, some of it may need to be reduced, summarized, or selected more carefully.

Embedding ​

An embedding is a numerical representation of text or other content that helps a system compare similarity between pieces of information. Embeddings are commonly used in retrieval and semantic search systems.

Fine-Tuning ​

Fine-tuning is the process of training a model further on additional examples so it behaves in a more specialized way. In privacy-focused systems, users often want assurance that their prompts and documents are not being used for fine-tuning.

Grounding ​

Grounding means giving the AI system reliable source material so its answer is based on real supporting information instead of only its general model knowledge.

Hallucination ​

A hallucination is an AI output that sounds correct but is unsupported, incorrect, or invented. Good instructions, relevant context, and careful review help reduce hallucinations.

Inference ​

Inference is the process of running a prompt through an AI model to generate an output. In user-facing tools, inference is what happens when you click a button to generate an answer, proposal, summary, or image.

Large Language Model (LLM) ​

A large language model, or LLM, is an AI model designed to understand and generate text. LLMs are commonly used for chat, summarization, question answering, drafting, and other language-based tasks.

Prompt ​

A prompt is the instruction or input sent to an AI system. A prompt may include the user's question, formatting instructions, document context, and project-specific guidance.

Prompt Engineering ​

Prompt engineering is the process of adjusting prompts to improve output quality, structure, tone, or detail. This can include changing wording, adding instructions, or clarifying constraints.

RAG (Retrieval-Augmented Generation) ​

RAG is a method that combines retrieval and generation. The system first retrieves relevant content from a document library or search layer, then supplies that content to the AI model so the generated response is better grounded in your own materials.

Retrieval ​

Retrieval is the process of finding relevant content from stored documents or other data sources before asking the AI model to generate a response.

Semantic search looks for content based on meaning rather than only exact keyword matches. This helps a system find relevant passages even when the wording is not identical to the search query.

Stateless Model ​

A stateless model processes each request independently rather than relying on long-term remembered state inside the model itself. In privacy-focused workflows, this helps reduce unnecessary persistence of user prompts and context.

Stored Completion ​

A stored completion is a generated output that is saved by an application after the model returns it. This is different from a model remembering the prompt. An application may save generated results or chat history in its own database even when the model itself is stateless.

Temperature ​

Temperature is a model setting that influences how predictable or creative the output is. Lower temperature usually produces more consistent output, while higher temperature can increase variety or creativity.

Token ​

A token is a unit of text a model uses internally when reading input and producing output. Token limits affect how much text can be processed in a single request.

Vector Database ​

A vector database is a storage system designed to hold embeddings and retrieve content based on similarity. Vector databases are often used in AI retrieval systems to help find document passages related to a question or prompt.

Platform Terms ​

Agent Briefing ​

An auto-generated summary of what the AI has learned about an opportunity from its source documents, such as response-length rules and submission requirements, shown on the project workspace.

Approved Answer ​

A vetted answer promoted for reuse. Approved answers are stored in the Approved Answers library and used to ground future responses. See Approved Answers.

The tool that searches public SAM.gov contract opportunities and uses AI to match them to your company profile. Matching opportunities can be turned into projects with Add to Project.

Chat ​

Privata's private, stateless conversational AI. You can ask questions, draft short content, and attach collections for grounded, cited answers. See Using Chat.

Collection ​

A grouped set of Knowledge Base documents and websites, organized by subject, product, service line, or opportunity type so the right context can be attached to the right project.

Company Profile Settings ​

The company-wide configuration (identity, offerings, GovCon profile, compliance and review policy) that powers opportunity matching, compliance checks, project labels, and the Approved Answers policy. See Company Profile Settings.

Company Workspace ​

The shared workspace every account belongs to. Projects, collections, and other assets are shared within the workspace, and it is administered by one or more Org Admins.

Confidence Score ​

The system's estimate of how well a generated Q&A answer is supported by your material. Confidence scores appear on Q&A answers and help you decide which need more review.

Connected Content ​

Files imported into collections from connected services (Google Drive, SharePoint, and Salesforce) without permanently storing the originals.

Essay Section ​

A narrative, solution-style section in the Response Editor that addresses requirements together as one flowing response, rather than item by item.

Extraction Agent ​

The AI step that reads an uploaded source document and pulls out its questions and requirements automatically for the Response Editor.

Extracted Instructions ​

Response requirements, constraints, or guidance pulled automatically from a source document and stored in the project instructions area.

Knowledge Base ​

The company's library of uploaded documents and websites that provides supporting context for search, Chat, summarization, and answer generation. It contains collections and an Approved Answers tab.

The feature that searches the Knowledge Base for relevant matches, ranks them by relevance, and lets you preview and open the results.

Labels ​

A controlled vocabulary describing what documents, collections, and projects are about (for example product/service, customer sector, compliance framework, document type). Labels focus retrieval so the AI uses the most relevant content.

Org Admin ​

A company administrator who can invite and manage users, manage billing and seats, and configure company-wide settings.

Project ​

The main workspace for one opportunity, bid, contract response, or proposal effort. It groups source documents, collections, tasks, instructions, and generated outputs.

Project Instructions ​

Stored responder guidance used to shape AI outputs. Instructions can be entered manually or extracted from the source document.

Project Labels ​

The labels applied to a project (Response Style, Products/Services & Offerings, Customer Sector, and Certifications) that focus retrieval and set the response style for that project.

Project Readiness Review ​

An analysis that checks whether a submission is complete (covering completeness, pricing, approvals, submission instructions, and compliance) and turns gaps into tasks. See Project Readiness.

Proposal Collateral Studio ​

The AI tool that turns your Knowledge Base into proposal collateral across several content types: presentations, company backgrounds, capability statements, financial profiles, past performance, certifications, team resumes, references, and technical approaches.

Q&A Section ​

A section in the Response Editor whose extracted questions or requirements are answered one at a time.

Required Indicator ​

A marker on an extracted question that appears to need a mandatory response, used to help prioritize review.

Response Editor ​

The workspace where you build a response from sections (Q&A, essay, and table), generate and refine answers, review, and download. Opened with Build Response. See Working with the Response Editor.

Response Style ​

A project label that sets the tone and format the AI uses when generating answers for that project.

Retrieval Scope ​

A project setting that controls whether the AI retrieves from only the collections listed on the project ("Listed only") or from all collections in the company Knowledge Base ("All collections").

Source Document ​

The uploaded file the platform processes as the main item to be answered: usually an RFP, RFI, SOW, or amendment.

Submission Readiness ​

The ledger on the project workspace that shows the current readiness verdict, individual checks (rungs), and how many have passed. Backed by the Project Readiness Review.

Supporting Documents ​

Files added to collections to help the AI generate grounded output: product guides, pricing sheets, company information, technical references, statements of work, and similar materials.

Table Section ​

A section in the Response Editor built from a table extracted from the source document, such as a pricing sheet or responsibility matrix. Generate table answers fills the cells, which you can then edit and save.

Task ​

A unit of work on a project, with an assignee and due date. Tasks can be created manually or automatically by the compliance and readiness checks. See Tasks & Collaboration.

For more detailed workflow examples, continue to the System Overview, Working with Projects, Building Your Knowledge Base, and Working with the Response Editor pages.