Creating an Agent

Agents are the cornerstone of Doti's intelligence layer - capable of not only retrieving data but executing reasoning steps, planning, and generating meaningful outputs. Creating an agent allows you to tailor Doti’s behavior for a specific goal, use case, or audience.


Getting Started

To create a new Agent, navigate to the Doti Agents Page.

⚠️ Only Doti Admins and Agent Creators have permissions to create and configure agents.


Agent Configuration Fields

When creating a new agent, you will configure the following:

1. Name & Icon

Choose a descriptive name and icon that represents the agent’s purpose. Example: Customer Prep Agent, Release Brief Agent, Support Debugger

Agent names and icons will also appear in Slack when the agent is tirggered!

2. Knowledge Collections

Assign the sources (collections) of knowledge the agent will use. These can include:

  • Slack Channels

  • Google Drive folders

  • Confluence Spaces

  • Salesforce Data

  • Gong Calls

  • Jira Projects

  • And more...

You can also create agents that search only within specific collections for high-control use cases.

3. Members

Assign access to this agent. Only selected users or teams will be able to interact with it via Slack or the browser extension.

New members of the agent will be notified about being added to it.


Instructions: The Heart of the Agent

Agent Instructions define what the agent does, how it does it, and how it communicates.

These aren’t just for tone — they are literal task blueprints.

They tell the agent:

  • What the objective is

  • Which data sources to prioritize

  • What the structure of the output should be

  • How to combine different pieces of information

  • What reasoning steps to follow


🔥 Example: “Customer Preparation Agent”

Here's an advanced example of an instruction set you can give your agent:


Your Role

You are a meeting preparation assistant. Your goal is to provide a concise and actionable summary for upcoming customer meetings. Use data from Slack, Google Drive, Salesforce, and Gong.

** Make sure to include data only related to the customer name provided. **

Output Structure

1. Customer Overview

  • Company Name

  • Industry

  • Key Contacts (with titles)

  • Recent News/Updates (past month)

  • ARR (from Salesforce, if exists)

2. Opportunities & Critical Cases

  • Open opportunities: Name, Amount, Stage

  • High-priority cases: Subject, Priority, Created Date

3. Critical Points to Address

4. Relationship Summary

  • Last 3 interactions (dates, people, discussion points)

  • Current Opportunities: Name, Stage, Value, Key Decision Makers

  • Outstanding Issues

5. Meeting Objectives

  • Customer Goals

  • Our Goals

  • Anticipated Objections

6. Gong Call Insights (if available)

  • Call Dates

  • Speakers

  • Sentiment (Positive, Neutral, Negative)

  • Action Items

7. Relevant Documents (from Google Drive)

  • Document Name

  • Description

  • Link

8. Slack Summary

  • Key topics from relevant Slack channels (last week)

  • Channel Name, Topics, Sentiment


Final Tips

  • You can edit instructions at any time - agents immediately adapt.

  • Combine agents with testbed autopilot to validate performance in Slack.

  • Use agent-scoped tokens to invoke the agent programmatically via API.

Last updated

Was this helpful?