Per-agent configuration
Each agent with its own instructions, goals and rules.
One agent to answer questions on the website, another to suggest replies to the team. Each with its own job, its own instructions and its own limits, without mixing them up.
- Nothing to install
- Implementation with a specialist
- Instructions with version history
- Test before publishing
Its own configuration
What is per-agent configuration?
It is giving each AI agent its own configuration: what it should achieve, how it talks, what it knows and what it must not write.
- 1
Goal
What the agent should achieve in each conversation.
- 2
Instructions and tone
How it talks and what it can do.
- 3
Files
What it checks to reply.
- 4
Safety rules
Filters for what it must not write.
The settings
What changes from one agent to another
- 01
Goals
Say what the agent should achieve in each conversation, for example, understand the request and collect the details for a quote.
- 02
Instructions and tone
You define how the agent talks and what it can do. Every change is saved, and you can go back to a previous version.
- Version 4 · saved
- Restore this version
- 03
Files
The files the agent checks to reply.
- 2026 catalog
- Returns policy
- Frequently asked questions
- 04
Safety rules
Filters for what the agent must not write, such as profanity, email addresses, phone numbers and sensitive numbers. The expert team turns on the right rules during implementation.
An agent that tries to handle everything ends up confusing the customer.
How to split
Who asks what, and which agent replies
Website visitors ask one thing; the people working in the inbox need another. Each agent, including the team assistant, counts toward the plan's agent limit.
Website agent
"Are you open on Saturdays?"
Replies with the hours and address that are in the files.
Team assistant
"How do I answer this discount request?"
A question from someone on the team. The assistant proposes the reply, and the person reviews it before sending.
Where each one works
Each agent where it makes sense
Website chat
The website agent replies through the code you paste on your website.
See the website chatInbox
The assistant that suggests replies sits in the conversation menu.
AI suggestions
Before it goes to work
Test each agent before putting it to work
Talk to the agent using the questions that really come in. Adjusted an instruction? Ask the same questions and compare with the previous version's reply, which stays saved.
- The goal shows in the conversation
- The tone is your company's
- The replies come from your files
- The rules filter what they should filter
- Every adjustment becomes a new version, and the previous one stays saved
Evaluation per agent
In the same account, each agent evaluated by its own goal
Each agent is evaluated by the goals you set for it, and each conversation gets a score, with what went well and what was missing.
Example
Website agent
Replies in: website chat.
- Evaluation on
Example
Quote page agent
Replies in: website chat.
- Evaluation on
An honest limit
Evaluation must be turned on and the goals set.
Questions
What people ask about having more than one agent
It depends on what you want from AI. One possible setup has two: the agent that replies in the website chat and the assistant that suggests replies to the team in the inbox. They are different jobs, with different instructions.
One agent for the website, another to help the team.
Build one agent for each job in your operation.
Create an account and build the first agent with the expert team.
Create account- Nothing to install
- Implementation with a specialist