Steps to Create a Custom Agent
From the Agents menu :
Click on the Agents page in your menu, then on "Create a new agent":
From a call :
On the call page, locate and click on the Ask Modjo AI tab, typically open by default.
Click on "@ Add an agent" then "Create my agent"
Write the instructions that will guide your agent's behavior. For a custom agent, specify:
The agent's main objective
The type of analysis or task to perform
The desired output format (list, paragraph, table)
Specific elements to search for or extract
Constraints and limits to respect
Select the Modjo data sources your agent will have access to. Currently, available sources include Modjo sources:
Call transcriptions: full text of recorded conversations
Conversation metadata: information about participants, duration, date
Deal information: data related to commercial opportunities associated with conversations
Contact data: information about contacts participating in conversations
Name your agent
Write a description for your agent
Click on
Save agentThe agent is created and becomes available in your library.
Once created, the agent will be accessible and editable by all users. However, only the agent's creator will be able to delete it.
Using Your Agent on Your Conversations
To use your newly created agent on other conversations:
Open another call page
Access the
Ask Modjo AItabSelect your agent from the list of available agents
Click on
Runor the equivalent to launch the analysis
Best Practices
Writing Instructions
Be specific: the more precise your instructions, the better the results
Provide examples: include examples of the expected output format
Define the context: specify the agent's role ("You are an assistant...", "You must...")
Limit the scope: focus each agent on a specific task rather than multiple ones
Add constraints: specify what the agent should NOT do if relevant
Naming and Organization
Use descriptive names that reflect the agent's function
Include the team name or use case if the agent is specific
Create multiple specialized agents rather than an "all-in-one" agent