Setting Up Agents

Setting Up Agents

An agent is an AI-powered team member that handles one specific job for your business. Think of each agent as a specialist: one qualifies leads, another answers support questions, another handles onboarding. Each agent has its own instructions, knowledge, and connected channels.

This guide walks you through creating your first agent, configuring it, and getting it live.

When to Use Agents

Agents work best for repetitive conversations that follow a pattern:

  • Qualifying inbound leads before they reach your sales team
  • Answering common product or pricing questions
  • Guiding new customers through setup
  • Handling support requests via chat or email
  • Keeping CRM records up to date from conversation data

If you find yourself (or your team) having the same conversation over and over, that is a good candidate for an agent.

Creating an Agent from a Template

  1. Go to AI > Agents in your dashboard
  2. Click Create Agent
  3. Choose a template that matches the job you need done
  4. Give your agent a name and configure its settings
  5. Connect it to a source (where conversations come from)
  6. Attach a knowledge base (so it knows about your products)
  7. Save and activate

Templates give you a fully configured workflow out of the box. You can customize everything after creation.

Choosing the Right Template

Each template is designed for a specific role. Pick the one closest to the job you need done, then adjust.

Lead Qualification

Best for: qualifying inbound website visitors before they reach your sales team.

The agent asks qualifying questions, captures contact details, scores the lead, and creates a CRM record. High-scoring leads get routed to your team immediately.

Pre-Sales

Best for: answering product questions during the evaluation phase.

Handles technical questions, explains how features work, compares plans, and helps prospects understand if your product is the right fit. Useful when buyers are researching but not ready to talk to sales yet.

Sales Closer

Best for: handling pricing discussions and moving deals forward.

Responds to pricing objections, explains ROI, walks through plan comparisons, and nudges prospects toward a decision. Works well when leads are qualified but stalling.

Onboarding

Best for: guiding new customers through their first steps.

Walks new users through account setup, connects integrations, explains key features, and answers getting-started questions. Reduces time-to-value and support tickets.

Customer Success

Best for: ongoing account management and retention.

Handles account questions, surfaces usage insights, identifies upsell opportunities, and flags at-risk accounts. Think of it as a proactive account manager that never sleeps.

Customer Service

Best for: general support across email and chat.

Triages incoming requests, answers common questions from your knowledge base, and escalates complex issues to your team. Works across both email and website chat.

Data Entry

Best for: keeping your CRM accurate without manual work.

Extracts information from conversations (job titles, company size, budget, timeline) and automatically updates CRM profiles. Runs in the background so your team never has to copy-paste between tools.

Configuring Your Agent

Once you have created an agent, there are four key settings to get right.

Name

Your agent's name is visible to visitors in chat. Choose something that fits your brand. Some teams use a human-sounding name ("Sarah from Acme"), others keep it clear ("Acme Support"). Either works - just be consistent.

Source

The source is where your agent listens for conversations. Connect it to:

  • Website chat widget - for live visitor conversations on your site
  • Email - for inbound email handling
  • Other channels - depending on your connected integrations

An agent needs at least one source to function. You can connect multiple sources if the same agent should handle conversations from different channels.

Knowledge Base

Without a knowledge base, your agent gives generic answers. With one, it knows your products, pricing, processes, and policies.

Go to AI > Knowledge to create a knowledge base, then attach it to your agent. Include:

  • Product documentation
  • Pricing details
  • FAQ content
  • Sales plays and objection handling guides
  • Process documents (how to book a demo, how to start a trial)

The more relevant content you provide, the better your agent performs.

Goals

Goals tell the agent what success looks like. Examples:

  • Qualify the lead and capture their email, company, and use case
  • Book a demo using your scheduling link
  • Resolve the support question without escalating
  • Collect feedback after resolving an issue

Clear goals keep agents focused. Without them, conversations tend to drift.

Human-in-the-Loop Review

When enabled, agent responses are held for your review before being sent. You can approve, edit, or reject each message.

This is useful when you are first setting up an agent and want to verify quality. Once you are confident in the responses, you can disable review and let the agent run autonomously.

Setting Up the Agent System Prompt

The agent system prompt is a workspace-level instruction that applies to all your agents. It tells every agent who your company is, what you sell, and how to behave.

To set it up, go to AI > Knowledge > Agent System Prompt.

What to Include

Keep it factual and specific:

  • Company name and description - what you do in one or two sentences
  • Products and services - what you sell, key features, who it is for
  • Pricing - plan names, price points, what is included in each
  • Common integrations - tools you work with
  • Tone of voice - formal, casual, technical, friendly

Identity Rules

Add rules that keep the agent consistent with your brand:

  • "Always refer to the company as 'we' and 'our', never 'they' or 'their'"
  • "Never mention that you are an AI unless directly asked"
  • "If asked about competitors, focus on our strengths rather than comparing"
  • "Always include a next step at the end of each response"

Example

You work for Acme Corp, a project management platform for construction teams.

Products:
- Acme Pro ($49/month): For small teams up to 10 users
- Acme Business ($149/month): For companies up to 50 users, includes reporting
- Acme Enterprise (custom pricing): Unlimited users, SSO, dedicated support

We integrate with AutoCAD, Procore, Sage, and QuickBooks.

Rules:
- Use a professional but friendly tone
- Always refer to Acme Corp as "we"
- If someone asks about pricing, recommend the plan that fits their team size
- Never promise features that are not listed above
- End every conversation with a clear next step

This prompt applies to all agents in your workspace. Individual agents add their own role-specific instructions on top of it.

Tips for Getting Started

  • Start with one agent. Get a single agent working well before adding more. It is easier to tune one agent than debug three at once.
  • Enable human-in-the-loop first. Review responses for the first week to catch issues early. Disable it once you trust the output.
  • Write a good system prompt. This is the single highest-impact thing you can do. A clear, factual system prompt with specific pricing, product details, and rules makes everything better.
  • Attach a knowledge base early. Even a few documents with your FAQ and pricing make a big difference compared to no knowledge base at all.
  • Check corrections regularly. When you correct an agent's response, it learns from that correction. The more you correct early on, the faster it improves.