Glossary

Definitions for the terms used throughout Lyzr documentation.

​

Agent

An agent is an LLM configured with a role, a goal, and instructions, plus optional connections to tools, knowledge, and memory. In Lyzr, each agent automatically exposes a persistent REST endpoint.

​

API (Application Programming Interface)

The Lyzr API is a set of RESTful endpoints that allow programmatic interaction with agents, tools, and services. All Lyzr API endpoints conform to the OpenAPI 3.1 standard.

​

AIMS (AI Management System)

AIMS is Lyzr’s console for monitoring, managing, and governing agent deployments, usage metrics, and audit logs.

​

Audit log

An audit log is a chronological record of events and actions taken by agents, users, or systems. It is used for compliance verification and debugging.

​

Bias mitigation

Bias mitigation refers to techniques and processes Lyzr applies to detect and reduce unwanted bias in AI outputs, helping ensure fair and consistent responses.

​

Chain of thought

Chain of thought is a prompting technique where intermediate reasoning steps are made explicit before a final answer. It improves model accuracy on multi-step problems.

​

Connector

A connector is a pre-built integration between Lyzr agents and an external service (such as Slack, Salesforce, or Google Sheets) that lets an agent fetch data or take actions.

​

DAG orchestration

DAG (Directed Acyclic Graph) orchestration is a workflow execution model where tasks run in a defined sequence or in parallel, based on explicit dependencies declared in a graph. In Lyzr, DAG orchestration is implemented through SuperFlow.

​

Data ingestion

Data ingestion is the process of importing documents, databases, or other data sources into Lyzr for indexing and retrieval.

​

Embedding

An embedding is a numeric vector representation of text (or other data) that captures semantic meaning. Embeddings are used to measure similarity between pieces of text and to power semantic search.

​

Embedding model

An embedding model is an algorithm that converts input data (such as text or images) into embeddings. Common examples include OpenAI’s text-embedding-3 and Cohere’s embed-v3.

​

Fine-tuning

Fine-tuning is the process of adapting a pre-trained LLM to a specific task or domain by training it on a smaller, task-focused dataset.

​

Generative AI

Generative AI refers to AI systems that produce new content — text, images, audio — based on patterns learned from training data.

​

Intent recognition

Intent recognition is the process by which an AI agent identifies the user’s goal from their input and selects the appropriate actions or responses.

​

JSON Schema

JSON Schema is a JSON-based format for defining the structure, required properties, and data types of JSON documents or API payloads. Lyzr uses JSON Schema for structured outputs and custom tool definitions.

​

Knowledge Base

A Knowledge Base is a repository of documents, web pages, or database content that agents query to ground their responses in real-world information. See also: Knowledge Graph, Semantic Model.

​

LLM (Large Language Model)

An LLM is a large neural network — such as GPT, Claude, or Gemini — trained to understand and generate human-like text. In Lyzr, agents run on top of LLMs and can be switched between providers at any time.

​

Managerial orchestration

Managerial orchestration is a dynamic orchestration mode where a manager agent decomposes an objective into subtasks and dispatches worker agents at runtime. The execution path varies based on the input.

​

Memory

Memory refers to mechanisms that allow an agent to retain and use information beyond the current turn. Lyzr supports short-term memory (per-session), long-term memory (summarized across turns), and cross-session memory via Cognis.

​

Model-agnostic pipeline

A model-agnostic pipeline is an architecture that lets you swap between different LLM providers — such as OpenAI GPT, Gemini, Claude, or Bedrock — without rebuilding the workflow. Lyzr agents are model-agnostic by default.

​

Multi-agent workflow

A multi-agent workflow is an automated process where multiple specialized agents interact or run in sequence to accomplish a task that is too complex for a single agent.

​

Orchestration

Orchestration is the coordination of multiple steps, tools, or agents into a single automated workflow. Lyzr provides two orchestration modes: Manager Agent (dynamic) and SuperFlow (deterministic DAG).

​

Parser

A parser is a component that processes and structures raw data — such as JSON or HTML — into a format suitable for downstream tasks or agent prompts.

​

Prompt

A prompt is the input text given to an LLM that guides its output. In Lyzr agents, the prompt is assembled from the role, goal, instructions, retrieved Knowledge Base context, memory, and the user’s message.

​

Prompt template

A prompt template is a reusable blueprint for prompts with placeholders that are filled dynamically at runtime with variables or data.

​

RAG (Retrieval-Augmented Generation)

RAG is a technique that combines information retrieval from external sources with generative model inference to produce responses grounded in specific documents or data.

​

Reranker

A reranker is a module that reorders retrieved results — documents or passages — based on relevance to the query before passing them to the generative model.

​

Role-Based Access Control (RBAC)

RBAC is a security model where permissions are assigned to roles rather than individual users. Lyzr uses RBAC to control which team members can build, deploy, or administer agents.

​

SDK (Software Development Kit)

The Lyzr SDK — also called the ADK, or Agent Development Kit — is a set of libraries and tools for building and managing agents programmatically in Python or TypeScript.

​

Agent Studio

Agent Studio is Lyzr’s visual interface for designing, testing, and deploying agents without writing code. It builds on the Agent Framework and exposes the full agent development lifecycle through a browser-based UI.

​

Tokenization

Tokenization is the process of breaking text into smaller units (tokens) that a language model processes. Token count determines the cost of an LLM call on most providers.

​

Vector store

A vector store is a database optimized for storing and querying vector embeddings for fast semantic search and similarity matching.

​

Webhook

A webhook is a user-defined HTTP callback URL that receives real-time notifications or data from Lyzr agents when specific events occur.

​

Zero-shot

Zero-shot inference is an inference mode where an LLM performs a task without any examples or prior fine-tuning, relying solely on its pre-trained knowledge and the current prompt.

​

Few-shot

Few-shot inference is an inference mode where an LLM is provided with a small number of examples in the prompt to guide its output for a specific task.

​

KPI (Key Performance Indicator)

A KPI is a measurable value used to evaluate the success of an agent or workflow against a business objective — for example, accuracy, response latency, or task completion rate.