AI Agent & LLM Engineering Glossary
Plain-English definitions of the terms that matter in agentic AI — written and maintained by the team behind our AI agent development services. Updated July 2026.
- AI Agent
- An AI agent is software that uses a large language model to autonomously plan and execute multi-step tasks — calling tools, querying data, and taking actions — rather than only generating text responses. Unlike a chatbot, an agent decides what to do next based on goals, context, and the results of its previous actions.Go deeper: How to build an AI agent from scratch →
- Agentic AI
- Agentic AI describes systems designed around autonomous action: models that plan, use tools, maintain state across steps, and pursue objectives with limited human intervention. It contrasts with generative AI used purely for single-shot content generation.Go deeper: How agentic AI is transforming SaaS →
- AutoGen
- AutoGen is an open-source framework from Microsoft for building multi-agent AI systems in which multiple LLM-powered agents converse with each other (and with humans) to solve tasks. It is strongest for research-style, conversation-driven agent collaboration and code-generation workflows.Go deeper: LangChain vs CrewAI vs AutoGen compared →
- CrewAI
- CrewAI is an open-source Python framework for building multi-agent AI systems organized as "crews" — teams of role-specialized agents (e.g., researcher, writer, reviewer) that collaborate on tasks with defined processes. It is known for the fastest path from idea to a structured agent team.Go deeper: CrewAI in production: real lessons →
- Guardrails
- Guardrails are the safety controls wrapped around an AI agent: input and output filtering, restricted action scopes, policy checks, and validation layers that prevent an agent from taking harmful, off-policy, or out-of-scope actions in production.
- Haystack
- Haystack is an open-source Python framework by deepset for building production LLM applications, particularly retrieval-augmented generation (RAG) pipelines and search systems. It provides composable pipeline components for retrieval, ranking, generation, and evaluation.
- Human-in-the-Loop (HITL)
- Human-in-the-loop is a design pattern where an AI agent pauses for human review or approval before executing high-stakes actions — sending money, signing contracts, contacting customers. It is the standard control for deploying agents in regulated or high-risk workflows.
- LangChain
- LangChain is the most widely adopted open-source framework for building LLM applications. It standardizes model access, tool calling, memory, and retrieval so developers can compose chains and agents across different model vendors without rewriting integrations.Go deeper: Long-term agent memory with LangChain →
- LangGraph
- LangGraph is LangChain’s library for building stateful agent workflows as graphs: nodes are steps or agents, edges define control flow, and built-in state persistence enables cycles, branching, retries, and human-in-the-loop checkpoints that linear chains cannot express.
- LLM Fine-Tuning
- Fine-tuning is the process of further training a pre-trained large language model on domain-specific examples so it adopts a desired style, format, or specialized knowledge. It is used when prompt engineering and RAG cannot reach the required accuracy — at the cost of training compute and maintenance.
- Model Context Protocol (MCP)
- The Model Context Protocol is an open standard (introduced by Anthropic) that lets AI agents connect to tools, data sources, and services through a common client-server interface. Instead of writing bespoke integrations per model vendor, teams expose systems once as MCP servers and any MCP-capable agent can use them.Go deeper: How AI agents use MCP →
- Multi-Agent System
- A multi-agent system coordinates several specialized AI agents — each with its own role, tools, and context — to complete work no single agent handles well. An orchestrator routes tasks, manages shared state, and resolves conflicts between agents.Go deeper: Multi-agent systems: enterprise guide →
- OpenAI Agents SDK
- The OpenAI Agents SDK is OpenAI’s framework for building agentic applications, providing primitives for agent handoffs, guardrails, and tracing. It succeeded the Assistants API as OpenAI’s recommended way to build production agents on its models.Go deeper: The Agents SDK explained →
- RAG (Retrieval-Augmented Generation)
- RAG is an architecture where an LLM’s answer is grounded in documents retrieved at query time from a trusted knowledge base — typically via a vector database — instead of relying only on training data. It is the standard technique for reducing hallucination and keeping AI systems current with private data.
- Semantic Kernel
- Semantic Kernel is Microsoft’s open-source SDK for integrating LLMs into .NET, Python, and Java applications. It provides plugins, planners, and memory abstractions, and is the common choice for agent development inside Microsoft-stack enterprises.
- Vector Database
- A vector database stores numerical embeddings of text, images, or other data and retrieves items by semantic similarity rather than keyword match. It is the retrieval layer behind RAG and agent memory — common choices include pgvector, Pinecone, Weaviate, and Qdrant.
Need These Concepts Working in Production?
We build custom AI agents on LangChain, CrewAI, and MCP — fixed-scope quotes in 48 hours.
Explore AI Agent Development
Address
India
First Floor, 208, 100 Feet Road, Mudaliarpet,
Puducherry, 605004, India.
Dubai
Arjumand Offices, Dubai Investment Park First,
Plot Number 645-0, UAE.
Ghana
Adama Akuredza Avenue Dzorwulu,
Accra, Ghana.
