Docs / AI Agents

AI Agents

Agents are AI assistants that can use tools and knowledge to answer questions, analyse documents, and collect structured data. Unlike a simple chatbot widget (which only does RAG search), an agent can call tools, run Python code, and search the web — all within a single conversation.

Agents vs Chatbot Widgets

A chatbot widget searches your collections and generates an answer. An agent does everything a widget does, plus it can execute code, search the web, and run multi-step workflows. Deploy an agent by linking it to a chatbot widget.

Creating an Agent

  1. Click Agents in the sidebar
  2. Click New Agent
  3. Fill in the Identity, Knowledge Sources, Tools, and Model sections
  4. Click Create Agent
  5. Click Publish on the agent detail page to deploy it to Azure AI Agent Service

Identity

Field Description
Name Display name shown in the chat header and the agent list
Description Short summary of what the agent does (visible to your team only)
Instructions System prompt that defines the agent's personality, knowledge, and rules. Be specific: mention the organisation name, the language to use, and any constraints.
Instructions matter

The instructions field is the single most important part of an agent. A well-written system prompt prevents hallucination, sets the tone, and keeps the agent on-topic. Start with: "You are a helpful assistant for [organisation name]. Only answer questions about [topic]. Always respond in [language]."

Knowledge Sources

Select the collections the agent can search. When a user asks a question, the agent retrieves relevant document chunks from these collections and uses them as context for its answer. This is RAG (Retrieval-Augmented Generation).

You can link multiple collections. The agent searches all of them for each query.

Tools

Tools extend what an agent can do beyond answering from documents:

Tool What it does
Knowledge Search Searches your linked collections using hybrid (semantic + keyword) search. Enable this to give the agent access to your documents. Requires at least one collection to be linked.
Code Interpreter Lets the agent write and execute Python code in a sandboxed environment. Use it for data analysis, calculations, chart generation, and processing uploaded files (PDF, DOCX).
Web Search Lets the agent search the web in real time using Bing. Use it when your documents may be out of date or when users ask about current events.

Model Settings

Setting Description
Model The Azure OpenAI deployment to use (e.g. gpt-4.1, gpt-4o-mini)
Temperature Controls response creativity. 0 = focused and deterministic (best for factual Q&A). 1 = creative. 2 = very random. Default: 1.

Publishing an Agent

Creating an agent saves the configuration but does not make it active. You must publish it to deploy it to Azure AI Agent Service:

  1. Open the agent detail page
  2. Click Publish (or Re-publish if the agent was already active)
  3. Wait a moment while aqoon provisions the agent on Azure
  4. The status badge changes to Active

After editing an agent's configuration, you must re-publish for the changes to take effect.

Status Meaning
Draft Agent created but not yet published
Provisioning Publishing in progress
Active Agent is live and ready to handle conversations
Error Publishing failed — check the error message and try again

Deploying via a Widget

An agent is deployed to end users through a chatbot widget. To link an agent to a widget:

  1. Go to Chatbot → open or create a widget
  2. In the widget's Agent field, select your published agent
  3. Save the widget — conversations through this widget now use the agent pipeline
Agent pipeline vs RAG pipeline

When an agent is linked to a widget, the agent handles all conversations using its configured tools and knowledge. When no agent is linked, the widget uses the simpler RAG pipeline (collection search + generate).

Agent Catalog

The Agent Catalog is a shared library of ready-to-use agent templates. You can install an agent from the catalog into your workspace with one click, then customise it for your needs.

Installing a Catalog Agent

  1. Click AgentsCatalog
  2. Browse available templates and click Use this
  3. A copy of the agent is created in your workspace
  4. Edit the instructions and settings to match your organisation, then publish

Publishing Your Agent to the Catalog

  1. Publish your agent (it must be Active)
  2. Open the agent detail page and scroll to the Catalog section
  3. Click Add to catalog — your agent is now listed and other organisations can copy it

To remove your agent from the catalog, click Remove from catalog in the same section. Other organisations that have already copied it keep their copy.

Deleting an Agent

Open the agent detail page, scroll to Danger Zone, and click Delete Agent. This removes the agent from Azure AI Agent Service and deletes all its configuration. Linked widgets fall back to the RAG pipeline. This action cannot be undone.