Setting Up Agents
This guide shows how to create an AI agent from a template and configure it for its job. Follow it when you deploy your first agent or add a new one.
Before you start
- Connect the source the agent will work on, such as a website source for chat or an email source. See chat widget and sources.
- Optionally, create a knowledge group with the content the agent should answer from. See knowledge bases.
Choose a template
Pick the template closest to the job. These are the customer-facing templates most teams start with:
| Template | What it does | Needs |
|---|---|---|
| Pre-Sales Agent | Answers product questions from website visitors using your knowledge base, and hands promising leads to a Lead Qualification Agent. | Website source |
| Lead Qualification Agent | Engages qualified leads with demo offers and onboarding guidance. Works with the Pre-Sales Agent. | Website source |
| Sales Closer | Handles pricing conversations, demo requests and purchase decisions. | Website source |
| Customer Service Agent | Drafts replies to inbound customer enquiries using your knowledge base. | Email or website source |
| Customer Success Agent | Supports customers through setup, ongoing support and product adoption. | Website source |
| Executive Assistant | Drafts email replies, schedules meetings and tracks commitments for one person. See Executive Assistant. | Email source |
| Data Entry Agent | Fills in visitor and contact profiles from conversations, such as company, role and team size. | None |
See AI agents for the full template list.
Create the agent
- In the sidebar, go to Automation > AI Agents and select Create Agent.
- Find a template. Browse the cards, or describe the job in What do you want this agent to do? and select Find Template. To build from scratch, select Start Blank.
- On the template card, select Use. If a card shows Connect Source instead, connect the source it needs first.
- In 1. Connect a Source, select the sources the agent should listen to. You can select more than one.
- In 2. Knowledge & Permissions, select a knowledge base and grant any permissions the agent needs, such as calendar access.
- Complete 3. Essentials if the template asks for required details.
- If the template shows 4. Personalization, complete it or select Skip for now.
- In Review & Deploy, check the summary and select Deploy Agent.
Outrun opens the new agent's page. It gives the agent a name, which visitors see in the chat widget. To change it, select Rename.
Configure the agent
On the agent's page:
- Agent Goals (Pre-Sales, Lead Qualification, Sales Closer, Customer Service and Customer Success agents): add up to 4 goals. Give each a Name, an Instruction and Success criteria, and optionally a Scheduling Link for booking meetings.
- Agent Instructions: add rules for this agent only, such as "Always ask how the customer wants proposals delivered."
- Knowledge: select the knowledge groups the agent searches when it responds.
- Reply to Sources: select the messaging sources the agent may reply on.
- Settings: turn human review on or off and choose a reviewer. See human-in-the-loop review.
Set the agent system prompt
The agent system prompt holds facts every agent in your workspace should know, such as your company name, products, plans and tone. Outrun adds it to every agent's instructions, and each agent's own instructions come after it.
- Go to Automation > AI Agents and open the Knowledge tab.
- Under Agent System Prompt, find Company Identity & Context and select Configure (or Edit).
- Write the prompt and select Save.
Example:
You work for Acme Pty Ltd, a project management platform for construction teams.
Products:
- Acme Team: for crews of up to 10 people.
- Acme Business: for firms with several sites.
We integrate with the accounting and design tools listed on our website.
Rules:
- Use a professional, friendly tone.
- Refer to Acme as "we".
- Never promise features that are not listed above.
- End every reply with a clear next step.
Tips
- Start with one agent. Get one agent working well before adding more.
- Keep review on at first. Review drafts to catch problems early, and turn review off only for agents whose drafts no longer need edits.
- Write a factual system prompt. Specific product, pricing and policy facts give agents less room to guess.
- Attach knowledge early. Even a small knowledge group with your FAQ and pricing gives the agent your answers instead of general ones.
- Check the Corrections tab. If the same correction keeps appearing, fix the agent's instructions or knowledge instead.