Glossary
AI Chatbot
A conversational program that understands and answers in natural language using a large language model, rather than following a fixed script. The current generation of chatbots, and the base layer for AI agents.
General definition
An AI Chatbot is a program that talks with people in ordinary language and generates its own replies with a large language model (LLM). That single difference separates it from the older, rule-based chatbot, which can only walk a user down a decision tree or match keywords to canned answers. An AI chatbot reads the whole conversation, works out what is being asked even when it is phrased badly, and composes an answer on the spot.
Under the hood most AI chatbots share the same four parts:
- A language model that predicts the words of a reply, hosted by a cloud provider or run on your own servers
- A system prompt that sets the persona, tone, scope and rules the model follows
- Retrieval (RAG) that pulls relevant passages from your documents into the prompt so answers rest on your data, not the model’s memory
- Tools the model may call, such as an order lookup or a booking API, which is the point at which a chatbot starts to become an AI agent
The line between an AI chatbot and an AI agent is autonomy. A chatbot answers the message in front of it; an agent is given a goal, plans several steps, calls tools, checks the results and carries on until the job is done. Most production deployments sit somewhere between the two. Typical uses: customer support and FAQ deflection, sales qualification on a website, internal knowledge assistants over policies and handbooks, and patient intake or appointment questions in healthcare.
The risks are as real as the benefits. Models can hallucinate confident but wrong answers; prompt injection can steer the bot with text hidden in a message or a document; and anything a user types can end up in a third-party model provider’s logs. The standard controls are grounding answers in retrieved sources, keeping secrets and permissions out of the prompt, moderating both input and output, choosing where the model runs, and giving users a clear route to a human.
In the Ethora ecosystem
The Ethora AI SDK treats an AI chatbot as an agent that lives inside a chat room or channel with its own profile, so it sits next to human participants rather than in a separate box. Each agent has its own system prompt, its own knowledge base (crawled website, uploaded files, per-agent live web indexing) for RAG, and its own model choice: a cloud model API or a self-hosted model reached through an OpenAI-compatible endpoint such as Ollama or vLLM, the self-hosted LLM path regulated teams tend to choose. Tool calling and the MCP server let the same agent act as well as answer.
On the safety side, messages in and out of an agent pass through the Trust and Safety layer, which can block, censor or quarantine content by stop words, regex or AI moderation and remove accidentally shared sensitive data. Everything the agent says is kept in the server-side archive with the audit trail, and the embeddable assistant widget puts the same chatbot on any website with a script tag. The older chatbot pattern and the wider conversational AI entry give the surrounding context; the AI SDK term explains the category.
