Skip to main content

​
Introduction

The Scalt Assistant is your copilot for creating intelligent agents. Rather than manually configuring each parameter, you simply describe what you want to automate, and the Assistant generates the agent for you.
Simple as a conversation. Powerful as code.

​
How does Scalt Assistant work?

Scalt Assistant uses artificial intelligence to:
  1. Understand your need expressed in natural language
  2. Analyze available tools and integrations
  3. Generate the optimal agent configuration
  4. Suggest improvements and optimizations

​
Access Scalt Assistant

1

Open your project

Select the project where you want to create the agent
2

Launch the Assistant

Select “Scalt Assistant”
3

Start the conversation

The Assistant interface opens, ready to receive your instructions

​
Write a good prompt

​
Basic principles

A good prompt for Scalt Assistant should be:

Clear

Clearly express the agent’s objective

Specific

Specify the actions to perform

Contextual

Give the necessary business context

Complete

Include triggers and expected results
Use this structure for better results: 
[OBJECTIVE]: What should the agent do?
[TRIGGER]: When should it act?
[ACTIONS]: What steps should it follow?
[RESULT]: What is the expected deliverable?

​
Prompt examples

​
Example 1: Customer support agent

Create an agent that responds to customer support emails
The more detailed your prompt, the more accurate and effective the generated agent will be.

​
Refine your agent with the Assistant

Scalt Assistant is conversational. You can refine your agent through multiple exchanges:
1

Initial prompt

Describe your general need
2

Assistant proposes

It generates a first version of the agent
3

You refine

“Also add spam verification” “Use GPT-4 model instead of Claude” “Send notifications to Teams rather than Slack”
4

Iteration

Assistant adjusts the configuration
5

Validation

Once satisfied, validate creation

​
Test your agent

Once created, immediately test your agent:
1

Access the Playground

Assistant automatically redirects you to the Playground
2

Simulate scenarios

Test with real data or examples
3

Check results

Review actions performed and results
4

Adjust if needed

Return to Assistant for modifications
Get inspired by these examples:
“Create an agent that sends a personalized welcome email to each new customer with advice based on their profile” Tools :
  • Send email (BREVO)
  • Get customer profile (CRM)
“Create an agent that monitors competitors’ prices on their website every day and alerts me if they lower their prices” Tools :
  • Web scraping (Firecrawl)
  • Send message (Slack)
“Create an agent that automates onboarding: creates accounts, sends access credentials, schedules a dem” Tools :
  • Create account (Auth0, Google, Slack)
  • Send message (Slack)
  • Schedule demo (Meet, calendar)
“Create an agent that generates and sends invoices on the 1st of the month, follows up on unpaid ones after 15 days, and updates accounting” Tools :
  • Get invoice (Zoho)
  • Send message (Outlook)
  • Schedule follow-up (Calendar)

​
Limitations and considerations

Points of attention:
  • Assistant generates a base configuration. For very complex needs, manual editing may be necessary.
  • Always verify permissions and access to third-party tools.
  • Test thoroughly before deploying to production.
  • Agents generate costs (Scalt credits + third-party APIs).

​
Best practices

Start simple

Create a basic version then enrich progressively

Test regularly

Verify each addition before complexifying

Document

Add comments for you and your team

Monitor

Check performance and costs regularly

​
Next steps

Edit your agent

Customize and optimize your agent

Run tasks

Execute and schedule your automations