AI Agent

General definition

AI agents extend the idea of a language model from “generate text” to “get things done.” An agent is given a goal and a set of tools - functions it can call to read data, write to systems, search the web, or communicate with other services. It plans a sequence of steps, executes them, observes the results, and adjusts until the goal is reached or it determines it cannot proceed.

• Perception - reading inputs from messages, documents, APIs or sensor data
• Reasoning - using an LLM to plan which actions to take and in what order
• Action - calling tools (web search, database queries, sending messages, running code)
• Memory - retaining information across steps via context or an external store such as a vector database

Agents can be single-step (one LLM call plus one tool) or multi-step loops that run until a stopping condition is met. They are described as agentic when they make consequential decisions autonomously. Frameworks such as MCP (Model Context Protocol) standardise how agents discover and call tools across different providers.

In the Ethora ecosystem

The Ethora AI SDK makes it straightforward to deploy AI agents that live inside your messaging product. An agent can join a group chat room, respond to user messages, look up information from a connected knowledge base, and take actions via tool calls, all while conversations stay on your own servers. The self-hosted LLM agent path is especially relevant for healthcare, finance and insurance teams where data cannot leave the organisation’s environment.

Agents built on Ethora benefit from the platform’s real-time messaging layer: they receive messages via WebSocket, can send rich replies, and can be scoped to specific apps or rooms. You supply the model (any OpenAI-compatible endpoint, a local Ollama instance, or a cloud provider) and Ethora handles the delivery, threading and persistence.