Agent Builder
An agent answers and acts in a chat. Give it a name, choose what it can do, and it is ready to talk: answering questions from your documents, browsing the web, reading PDFs, calling your API or running tasks on a schedule.
Creating an agent
Open Agents in the sidebar. You can start in two ways.
Create an agent
Click Create an agent (or + next to Agents at the top of the list) and fill in:
| Field | What to enter |
|---|---|
| Name | What the agent is called. It greets people with it: "Hi! I'm Support. How can I help you today?" |
| Purpose | What it is for, in a sentence or two. This becomes its instructions |
| Model | Let Ubex choose a model automatically, or pick one yourself |
| Level | Lite, High or Max: what the agent is allowed to do (see below) |
The agent opens straight away, ready to chat.
Start from what it should do
Under the button, three cards start an agent at a level that fits a typical job. Click Start with Lite, Start with High or Start with Max.
Levels: what an agent can do
| Level | In short | Can do | Example request |
|---|---|---|---|
| Lite | Conversational | Chat and answer questions | "Help me reply to this email" |
| High | Reads and creates content | Browse web pages, read PDFs and documents, see attached images, generate PDFs | "Read this PDF and list the key dates" |
| Max | Everything, plus code and APIs | Everything in High, call external APIs, run code | "Pull today's orders from our API" |
Pick the lowest level that does the job. You can change the level later in the builder.
The agent's canvas
Every agent is a small workflow you can open and change. Its canvas has three nodes:
- Chat Start - the chat trigger. Each message someone sends starts a run.
- Agent - the agent itself: its model, its instructions and the tools it may use. Its answer comes out as
agent_response. - Send Response - sends
{{agent_response}}back to the chat.
Click the Agent node to change its instructions, model and settings. Three switches there are worth knowing:
- Session Memory - remember earlier messages in the same conversation
- Scheduled Tasks - let people ask the agent to do something later or on a schedule: "every morning at 9, check our prices"
- Can call other agents - let it ask other agents in the workspace for help
See the Agent node for every field.
Adding tools
Tools go inside the Agent node. Double-click inside it and pick one from the list. The agent decides by itself when to use each tool. The ones used most:
- Web Search - look things up on the web
- Browser - open, read and fill in pages in a real Chrome that keeps logins
- Memory - remember things across conversations
- MCP Server - use an app you connected in Apps, such as GitHub, Notion or Slack
- Query Data - search your tables and documents in My Data
- HTTP Request, Code, Mail and more
Knowledge
To let an agent answer from your own material, add it to My Data first (upload PDFs, add a website or paste text), then put a Query Data node inside the agent that searches it.
Chat and Ask Vira
The panel at the bottom left of the canvas has two tabs:
- Chat - talk to your agent to test it. Use the microphone to speak, the trash icon to start over.
- Ask Vira - ask Vira to change the agent for you: "add web search", "answer in Spanish when the customer writes in Spanish", "keep answers under 80 words".
Active and Publish
At the top right:
- Active turns the agent on or off.
- History lists earlier versions; open one to go back to it.
- Publish puts your changes live.
Every conversation with the agent is a run, so you can follow them in Activity.
Next steps
- Agent node - every setting of the agent
- Chat Trigger - how messages start a run
- Apps - connect the tools your agents use