Agents Overview

Build intelligent AI agents with visual flow builder, toolsets, and collections

🤖

PipesHub Agents

Create custom AI agents that can perform actions across your connected applications

Beta

Visual Builder

What are Agents?

Agents are customizable AI assistants that can interact with your connected applications, search collections, and perform actions on your behalf. Unlike the standard chatbot assistant, agents are purpose-built for specific workflows and can be configured with:

• Reasoning Models - Choose the AI model with reasoning capabilities that powers your agent
• Toolsets - Connect to external services like Slack, Jira, Confluence, and more
• Knowledge sources - Access indexed documents and collections
• Custom system prompts - Define the agent’s personality and behavior

When you want to use both Connector and Toolset: In the Agent Builder add nodes from Knowledge (Collections, API Apps) for document search and context retrieval, and add nodes from Tools for performing actions. Connect both to your Agent Core so the agent can search context and take action in the same conversation.

Key Concepts

Visual Flow Builder

Agents are built using a visual drag-and-drop flow builder powered by ReactFlow. You create agents by:

1. Adding an Agent Core node that defines the agent’s basic configuration
2. Connecting Reasoning Model nodes (LLM with reasoning enabled) to power the agent’s intelligence
3. Adding Toolset nodes to give the agent access to external applications
4. Connecting Collection nodes for indexed document search
5. Configuring edges to connect all components together

Agent Core

The Agent Core is the central configuration for your agent:

Field	Description
Name	Display name for your agent
Description	Brief description of what the agent does
Start Message	Initial greeting when users start a conversation
Instruction	Detailed instructions for the agent’s behavior and capabilities
System Prompt	Instructions that define the agent’s behavior and personality

Toolsets

Toolsets are collections of tools that allow agents to interact with external applications. Each toolset provides specific actions:

• Jira Toolset - Create issues, search tickets, add comments, etc.
• Slack Toolset - Send messages, search channels, manage conversations, etc.
• Confluence Toolset - Create/update pages, search content, etc.
• Gmail Toolset - Send emails, read messages, search, etc.
• Calendar Toolset - Create events, update events, check availability, etc.
• Drive Toolset - List files, manage folders, search documents, etc.

Toolsets must be configured and authenticated before they can be used in agents. See Toolsets Overview for setup instructions.

Collections

Agents can access your indexed knowledge sources:

• Connectors - Data synced and indexed from connected applications for querying and search (Google Drive, SharePoint, etc.)
• Collections - Custom document collections you’ve uploaded

Agent Capabilities

What Agents Can Do

Capability	Description
Search Knowledge	Query indexed documents and return relevant information
Perform Actions	Create tickets, send messages, update pages via toolsets
Multi-tool Workflows	Combine multiple toolsets in a single conversation
Context-Aware Responses	Use system prompts to maintain consistent behavior

Getting Started

1. Configure Toolsets - Set up the toolsets your agent will use (Toolsets Overview)
2. Create an Agent - Use the visual builder to create your agent (Setup Guide)
3. Test and Iterate - Chat with your agent and refine its configuration

---

Frequently Asked Questions

What's the difference between Connectors and Toolsets?

Aspect	Connectors	Toolsets
Purpose	Sync and index data for search	Enable agents to perform actions
Data Flow	One-way (import data into PipesHub)	Two-way (read and write via API)
When to Use	Query/search data	Perform actions on data

Example: Use a Gmail Connector to index and search your email history. Use a Gmail Toolset to send new emails through an agent.

When should I use a Connector vs a Toolset?

Connector — Use when you need document context search: connectors sync and index your documents, emails, and pages so your assistant can find and use that context.

Toolset — When you need actions: create or update content, send messages, or run analytical queries (e.g. JQL, CQL, reports).

Quick takeaway: Search & context → Connector. Actions → Toolset.

Can I use one OAuth app for both Connector and Toolset?

Yes! You can create a single OAuth app in the service’s developer console and use it for both:

What you need to add in your OAuth app:

1. Both redirect URLs:

   • Connector callback URL (e.g., /connectors/oauth/callback/Gmail)
   • Toolset callback URL (e.g., /api/oauth/callback/Gmail)

2. Both scope types:

   • Read-only scopes for the Connector (to index data)
   • Read/write/delete scopes for the Toolset (to perform actions)

Benefits:

• Simpler setup - One OAuth app instead of two
• Same Client ID and Client Secret for both configurations
• Easier management in the service’s developer console

How do I know which nodes to connect in Agent Builder?

Follow this pattern:

1. Agent Core (required) - The center of your agent
2. Reasoning Model (required) - Connect at least one AI model
3. Toolset nodes (optional) - For actions like sending emails, creating tickets
4. Collection nodes (optional) - For searching indexed data from Connectors

All nodes must connect to the Agent Core. The agent will have access to all connected toolsets and collections.

Can I choose what knowledge source or tool the agent uses during chat?

Yes! While chatting with an agent, you can select:

• Specific Collections - Query data from particular Connectors (e.g., search only Confluence or only Drive)
• Specific Toolsets - Use particular tools (e.g., send via Gmail or create Jira ticket)
• Reasoning Model - Choose which AI model to use for the query

This gives you control at any time during the conversation to specify where the agent should search or which action to perform.

Can I connect multiple Reasoning Models to one agent?

Yes! You can connect multiple Reasoning Model nodes to the Agent Core.

When chatting with the agent, users can select which model to use for specific queries from the chat interface. This allows:

• Using different models for different complexity levels
• Flexibility to choose the best model for each task

---

Next Steps

Setup Guide

Step-by-step instructions for creating your first agent

Toolsets Overview

Learn about available toolsets and how to configure them