Building an agent with the AI builder (Stan) Guide

The fastest way to create a Praxivara agent is to describe what you want it to do in plain English. Stan, the AI builder on each agent's Build tab, reads your description and wires up the capabilities the agent needs — the tools it uses, the skills it follows, the knowledge it reads, and how it keeps you posted.

How the AI builder works

Instead of assembling an agent piece by piece, you have a conversation with Stan. You explain the job in your own words, and Stan translates that into a working configuration: it picks the right built-in tools, installs any expert skills that fit, and sets up how the agent reports back to you.

You stay in control the whole time. Stan proposes what it's setting up as it goes, and you can refine anything by simply telling it what to change.

Tip: Describe the outcome you want, not the mechanics. "Watch my brand mentions and email me a weekly summary" gives Stan enough to choose the right capabilities on your behalf.

Build an agent step by step

1. Open the Build tab on a new or existing agent to start talking with Stan.
2. Describe the job in plain English. Say what the agent is for, what it should do, and what a good result looks like. Include specifics like the tone, the format of the output, or how often it should run.
3. Let Stan set up the capabilities. As you talk, it wires up the tools and skills the agent needs and configures how it will notify you.
4. Add any details Stan asks for. If the agent needs to reach an outside service or connect an account, Stan will prompt you for what it needs.
5. Refine by asking. Want the agent to also send a WhatsApp message, follow a proven playbook, or change its tone? Just tell Stan, and it updates the setup.

What Stan wires up for you

A single plain-English description can set up several parts of your agent at once:

• Tools and integrations — the things the agent can actually do, from searching the web and working with files to using a connected account like Gmail or Slack.
• Skills — packaged, step-by-step playbooks for a specific job. Adding a skill also brings along the extra tools that skill needs, so the agent gets both the method and the means in one step.
• Knowledge — reference material the agent reads on every run so its output stays accurate and on-brand.
• Channels and notifications — how the agent keeps you posted, such as emailing a summary or pinging you on a messaging channel.

You can review everything Stan set up on the tabs under Configure on the agent's detail page.

Review and refine what was built

After building, open the Configure tabs to see the result. The Tools tab shows every tool and integration the agent has, grouped by integration with the connected account shown. The Skills tab lists the playbooks it follows. You can manage Knowledge and Secrets directly on their tabs, while other changes are best made by asking Stan.

Test before you rely on it

Use a test run to confirm the agent behaves the way you expect. If something's off — it skipped a step or lacked a capability — go back to the Build tab and tell Stan what to adjust.

Roll back if needed

Every change is tracked in the agent's version history, so you can compare versions and roll back to an earlier setup if a change didn't work out.

When the agent can't do something

If an agent doesn't complete a task, it's often missing the right capability. The most common fixes:

• It lacked a tool. Ask Stan to give it the capability — for example, "let it send WhatsApp messages."
• An integration isn't connected yet. An integration can appear on the Tools tab but still need an account connected before the agent can use it.
• It's not following a proven method. Ask Stan to add the matching skill.
• It's missing key facts. Add the relevant material to the agent's Knowledge so it reads it on every run.

Note: Built-in tools are always available to your agents, while integrations require connecting an account first. If you ever want a capability the agent doesn't have, describing it to Stan is the quickest way to add it.