Skip to main content
Build a custom AI Agent tailored to your team’s needs — whether from a plain-language description, a template, or a blank slate. Once created, you can configure every aspect of your Agent manually or use the Agent Copilot to do it with plain language.

Create a new Agent

1

Open the Agent creation screen

Navigate to the Agents page and click + New Agent in the top right corner.
Agent creation screen showing Custom Agent input, Create Blank Agent, Import Agent, and template gallery
2

Choose a creation method

Select one of four options:
  • Custom Agent — Describe your desired Agent in plain language and Cassidy generates it based on your description.
  • Create Blank Agent — Start with an empty Agent for full manual customization.
  • Import Agent — Upload an exported Agent JSON file. See Import and export Agents for details.
  • Template — Browse 75+ templates organized by common use cases like sales, support, and marketing.
3

Refine with the Agent Copilot

If you selected Custom Agent or a Template, the Agent Copilot opens automatically. Describe what you want your Agent to do, answer a few clarifying questions, and the Copilot builds out the instructions, name, description, and configuration for you.
Agent Copilot conversation during initial Agent creation
4

Review and customize

After creation, review and fine-tune your Agent’s settings in the editor. You can adjust the AI model, instructions, knowledge, connectors, capabilities, and brand voice at any time.
Agent editor showing Setup tab with Copilot panel displaying changes summary
5

Publish your Agent

When you’re satisfied with the configuration, click Publish to make the Agent available for your team. See Publish versions and review history for details on versioning and publishing.
Publish button dropdown in the Agent editor
Once published, your Agent is ready to use in Chat and Workflows.

The Agent editor

The Agent editor is organized into three tabs:
  • Setup — The core configuration for your Agent: its identity, AI model, instructions, knowledge, tools, and brand voice. This is where you define what your Agent knows and what it can do.
  • Deployments — Where you make your Agent available outside of the Cassidy app — in Slack, Microsoft Teams, Google Chat, embedded on your website, or via API. See Deploy your Agent for details.
  • History — A log of past conversations users have had with this Agent. Use it to review how the Agent is performing and identify areas to improve.
You can configure your Agent in two ways:
  • Manually — Edit each setting directly in the editor. The sections below walk through every option available in the Setup tab.
  • With the Agent Copilot — Describe what you want in plain language and the Copilot updates the configuration for you. You can use the Copilot to build from scratch, make targeted changes, or optimize your existing instructions.
Both approaches work together — use the Copilot to get started quickly, then fine-tune individual settings manually.

Configure your Agent

Appearance

Set your Agent’s icon, name, and description. The description helps your team understand what the Agent does and helps the system match the right Agent to a request.
Agent appearance section with icon, name, and description fields

AI model

Select the foundation model your Agent uses. Cassidy supports models from OpenAI, Anthropic, and Google, each with different strengths. See Choose the right AI model for a comparison.
AI model dropdown showing available foundation models
Expand Advanced Model Settings to adjust the temperature — a value that controls how creative or deterministic the Agent’s responses are.
  • Lower temperature (closer to 0): More consistent, predictable outputs. Best for tasks where the Agent should follow instructions precisely, like data extraction or compliance responses.
  • Higher temperature (closer to 1): More varied, creative outputs. Best for brainstorming, marketing copy, or other creative tasks.
Advanced Model Settings panel with temperature slider

Instructions

Write clear instructions that describe what the Agent should do, how it should respond, and any rules it should follow. Be specific — the more detail you provide, the better the Agent performs.
Instructions field showing structured instructions with role, objectives, and Knowledge Base references
Reference Knowledge Base items directly in instructions by typing # to search for and select documents. The Agent autonomously searches over referenced content when relevant.
Knowledge Base reference search popup appearing after typing # in the instructions field
You can pin a referenced item by hovering over it and clicking the pin icon. Pinned content is included directly in the prompt every time (counting toward token limits), so the Agent always sees it in full.
Pin icon on a referenced Knowledge Base item in the instructions field
Click Optimize at the bottom of the instructions field to have the Agent Copilot analyze and improve your instructions automatically.
Optimize button at the bottom of the instructions field
Agent instructions are one layer of AI customization. You can also set user-level personalization and organization-wide global instructions. See Custom AI instructions for details.

Knowledge

Connect your Agent to your Knowledge Base so it can search company documents and answer questions with accurate, cited responses. You can add entire collections, reference specific files in your instructions, and control when the Agent searches. Click + Add in the Knowledge section to browse and select sources.
Add Knowledge dialog showing Knowledge Base collections with checkboxes to select sources
For the full walkthrough — including Smart Search, inline references, and pinning — see Agent knowledge.

How Agents use tools

When you give an Agent tools — Connectors, Capabilities, and Workflows — you’re not programming it to follow a rigid script. Instead, the Agent decides on its own which tools to use based on the conversation. Think of it like giving a team member access to various resources. If someone asks a question about a customer, the Agent recognizes it should check Salesforce. If they ask for a chart, it reaches for Data Analysis. If they need the latest industry news, it uses Web Search. The Agent reads the user’s request, evaluates which tools are relevant, and calls them automatically — no manual triggering required. This means a single Agent can handle a wide range of requests as long as it has the right tools enabled. A sales Agent with Knowledge Base access, a HubSpot Connector, Web Search, and a follow-up email Workflow can answer product questions, pull up deal data, research a prospect’s company, and send a personalized email — all in the same conversation. You control what’s available by choosing which tools to enable. The Agent handles the rest.

Connectors

Connectors let your Agent interact with external business tools in real time. Unlike Knowledge Base integrations that sync content into Cassidy, Connectors query live data directly — nothing is copied or stored. Your Agent can look up a contact in HubSpot, run a Snowflake query, update a Salesforce record, or create a ServiceNow ticket, all within a conversation. Click + Add in the Connectors section to browse available connectors. Each connector can be configured with per-user or shared authentication and granular action permissions — you decide whether the Agent can read only, read and write, or whether write actions need user approval first.
Connectors section showing available connector options
Available connectors include HubSpot, Salesforce, Airtable, Snowflake, and more. See Overview of Connectors for the full list, connection methods, and permission controls.

Capabilities

Capabilities are built-in skills that extend what your Agent can do beyond conversation. They give Agents native abilities that don’t require an external integration to set up. Click + Add Capabilities to enable any of the following:
  • Data Analysis — Upload spreadsheets and CSVs and let the Agent analyze, interpret, and visualize your data. It writes and runs Python code behind the scenes to produce charts, calculations, and insights.
  • Web Search — Search the internet for current information during conversations. Useful for research, competitive intelligence, and answering questions that require up-to-date knowledge.
  • Image Generation — Create and edit images from text descriptions using AI image models.
  • Artifacts — Build interactive content like dashboards, calculators, reports, and prototypes that users can interact with directly in the chat.
  • Browser Use — Navigate and interact with web pages to complete tasks like filling out forms, extracting content, or taking screenshots.
Capabilities dropdown showing Data Analysis, Web Search, Image Generation, Artifacts, and Browser Use
See Overview of Capabilities for setup details and how each one works.

Workflows

Connect Workflows your Agent can trigger during a conversation. When a user’s request matches a connected Workflow, the Agent collects the required inputs and runs the Workflow — sending emails, updating records, generating documents, or any other multi-step automation. Click + Add Workflow to select from your published Workflows.
Workflows section in the Agent editor with an added Workflow
Add clear, descriptive names and descriptions to your Workflows so the Agent knows when to call them. The Agent uses the Workflow name and description to decide which Workflow matches a user’s request.
See Call Workflow for the full setup and how it works in practice.

Brand voice

Fine-tune your Agent’s writing style by uploading examples of your organization’s content. Click Analyze to have Cassidy study the examples and calibrate the Agent’s tone, vocabulary, and style so every response sounds like your company.
Brand voice section with the Analyze button
Brand voice analysis showing detected writing style characteristics

Use the Agent Copilot

The Agent Copilot is an AI-powered building partner that helps you configure Agents using plain language. Instead of editing each setting manually, describe what you want and the Copilot generates or updates the Agent’s configuration for you — including instructions, name, description, tools, and more. You can use the Copilot at any point: when creating a new Agent from scratch, when starting from a template, or when refining an existing Agent. It’s the fastest way to go from an idea to a working Agent.
Do not manually edit the Agent while the Copilot is working — your changes will be overwritten.

Access the Copilot

  • When creating a new Agent: The Copilot opens automatically if you choose Custom Agent or a Template.
  • When editing an existing Agent: Click the wand icon in the top toolbar of the Agent editor.
Wand icon in the Agent editor toolbar for accessing the Copilot

How the Copilot works

1

Describe what you want

Type a plain-language description of the Agent or the change you want to make. The Copilot plans its approach and begins building.
Copilot showing its planning process after receiving a description
2

Answer clarifying questions

The Copilot may ask follow-up questions to refine the Agent. Questions come as multiple choice, multi-select, free-text, or Knowledge Base attachment prompts. You can answer or click Skip to move on.
Copilot asking a multiple choice question to refine the Agent
3

Review the changes

Expand the changes summary to see which fields were modified — icon, name, description, instructions, and more. The Copilot explains what it did and why.
Copilot changes summary showing modified fields like Icon, Name, Description, and Instructions
4

Continue iterating

Send follow-up messages to refine further. You can ask the Copilot to update instructions, add tools, adjust the description, or make any other changes.
Copilot conversation showing follow-up refinements to the Agent

Copilot controls

ControlWhat it does
RegenerateTry a different approach for any Copilot message
Edit messageModify your prompt and resubmit
Thumbs up/downGive feedback on Copilot responses
Restore checkpointRevert the Agent to the version before a specific change
New chatStart a fresh Copilot conversation
Chat historyBrowse and return to earlier Copilot conversations

Custom AI instructions

Beyond the instructions you write directly on each Agent, Cassidy supports two additional layers of AI customization:
  • User-level personalization — Individual users can set their name, role, and custom instructions so the AI tailors responses in Chat. These apply across all chats regardless of which Agent is selected. See Personalize your chat for setup instructions.
  • Global instructions — Admins can set organization-wide rules that apply to every Agent (and optionally Workflows). Use these for brand voice, compliance guardrails, or company context.

Preview and test

Click the preview icon in the top-right corner of the Agent editor to open a test chat. This lets you try out your Agent and iterate on changes before publishing.
Preview chat panel in the Agent editor for testing the Agent

Review chat history

The History tab shows every conversation users have had with your Agent. Use it to monitor how the Agent handles real requests, spot patterns in what users are asking, and identify areas where the instructions or knowledge need refinement.
History tab showing a list of past conversations with the Agent including chat titles, users, and dates

Manage versions

Access version history from the Draft dropdown at the top of the Agent editor. From here you can save drafts, publish new versions, compare changes between versions, and revert to a previous version.
Draft dropdown showing version history with published versions, changes summary, and options to Publish Current Draft or Save Without Publishing
For the full walkthrough, see Publish versions and review history.

Deploy your Agent

Once published, your Agent is available in Cassidy’s Chat by default. To make it available where your team already works, open the Deployments tab in the Agent editor and configure one or more deployment channels:

Slack

Chat with your Agent directly in Slack channels and DMs.

Microsoft Teams

Add your Agent as a Teams bot for conversations and commands.

Google Chat

Deploy your Agent into Google Chat spaces.

Embed on your website

Add a chat widget to your website for customers or internal users.

API

Integrate your Agent into any application via the Cassidy API.

Next steps

Agent knowledge

Set up which Knowledge Base sources your Agent can search.

Share and manage Agents

Organize Agents into folders, manage permissions, and version history.

Chatting with an Agent

Learn about Chat features including personalization, file uploads, and voice mode.

Configure global instructions

Set organization-wide AI behavior rules for every Agent.