Tools¶

Configure tools that extend your agents' capabilities — from knowledge retrieval and CRM integrations to human handoff and custom API calls.

Overview¶

Tools are actions and integrations that agents can invoke during conversations. They allow your agent to go beyond simple text responses — retrieving documents, calling external APIs, handing off to human agents, creating CRM leads, and more.

Tools are configured at the workspace level under Settings > Processors > Tools, then assigned to individual agents under Agent Settings > Tools.

Available Tool Types¶

Core Tools¶

Context¶

The Context tool provides the agent with background information to work with. There are two primary types:

Static Context — fixed information that does not change:

Dynamic Context — obtained from the webpage the agent is operating on (Floating Widget/Sidebar only):

• Enable "Use Dynamic Context from the Webpage for Floating Widget/Sidebar Agent"
• Configure the Get Contents Code — a JavaScript snippet that extracts content from the page
• Default: return document.body.innerTEXT; (uses all page content)

Retrieval¶

The Retrieval tool performs vector search against a knowledge repository to find and return the most relevant document chunks for each user query. This is the core tool for RAG (Retrieval-Augmented Generation) workflows.

Chat with User Selected Documents¶

Enables end users to select specific documents from search results within a Search Type agent and initiate a focused chat conversation using only those documents as context.

Setup requires two agents:

1. Search Agent — enable "Enable Chat with Selected Documents" in Behavior settings and designate a Chat Type agent
2. Chat Type Agent — must be configured with the "Chat with User Selected Documents" tool, which inherits the parent Search Agent's knowledge repository

End-user flow:

1. User searches in Non-AI mode and selects up to 10 documents
2. Clicks "Start Chat" to open a new chat session with those documents as context — opens in the same browser tab (browser back button returns to the Search Agent view). Works consistently across Chrome, Firefox, Edge and Safari
3. Can toggle between Full Documents and Relevant Chunks modes
4. A real-time capacity visualization bar tracks token usage
5. System automatically falls back to Retrieval Mode if documents exceed 100% context capacity

Human Agent¶

Hands off the conversation from the AI to a human agent. Includes queue management, timeout configuration, and end-of-session rating.

Key configuration fields:

Agent to Agent¶

Communicate with another agent using API calls. Useful for specialized workflows where different agents handle different types of tasks. Cyclic loops are automatically prevented.

User Memory¶

Persists user-specific information across all conversations for a specific user. The agent can store and recall facts about the user (preferences, past interactions, account details) between sessions.

Tool Configuration Fields¶

User Memory tool configuration with the Merge Instructions field — rule-based merge logic (conflict_resolution, deduplication, …) is editable at the tool level; click Default to restore the built-in rules

Context Memory¶

Stores and retrieves contextual data with smart merging logic. Supports multiple persistence scopes:

Advanced configuration:

Integration Tools¶

Custom¶

A generic tool that executes a custom JavaScript function to perform any logic — calculations, data transformations, API calls, or custom business rules.

Example — Calculation tool:

Schema properties: operation, left, right

try {
  const operation = schema.operation;
  const left = parseFloat(schema.left);
  const right = parseFloat(schema.right);
  // Perform calculation based on operation...
}

Webhook¶

An HTTP tool for calling external APIs. The API must respond within 10 seconds.

Available placeholders in URL, Body, and Headers:

Additional Parameters:

Multi-Step Custom¶

A specialized tool that executes a series of steps in order — like a purchase journey or multi-stage workflow. Each step is a separate API call or action that can use outputs from previous steps.

Click "Add Step" to add more steps to the sequence. Each step executes in order, and outputs from earlier steps can be referenced by later ones.

Salesforce Create Lead¶

Tool for interacting with Salesforce APIs, focusing on lead management. Creates leads directly in Salesforce from conversation data.

Hubspot Create Lead¶

Tool for interacting with HubSpot APIs, focusing on lead management. Works identically to the Salesforce tool but maps to HubSpot's API structure.

Account Connection Mode — Global vs User¶

Both Salesforce and HubSpot lead creation tools support a per-tool Account Connection Mode, letting admins choose how authentication is handled:

Salesforce Create Lead tool with Account Connection Mode set to Global — the admin-configured account (Authorized Username, Connected/Disconnect controls) is shared across all users. Helper text below the toggle summarizes both modes Salesforce in User mode — no shared admin credentials. An Instance URL field is required so each user's credentials resolve to the correct Salesforce org (e.g., mycompany.my.salesforce.com)

HubSpot in User mode — no Instance URL required; the form goes straight from the mode toggle to the Schema

User Mode — End-User Connection Persistence

When a user connects their own credentials in User mode, the connection is persisted on the user's Account page under the User Connections section, where they can review or delete the connected account at any time. The table shows the Application (HubSpot / Salesforce), the originating Tool, Instance URL (when applicable), Username, Status, and a delete action.

End-user Account page — User Connections section listing a connected HubSpot tool (Status: Connected). The trash icon disconnects the account

Calendly¶

Tool for scheduling meetings directly within agent conversations. When invoked, the agent displays a Calendly scheduling interface (inline or popup) to the user.

Additional Parameters:

Data Visualization¶

A built-in tool type that renders interactive HTML visualizations directly in the chat. When invoked, the LLM authors a complete HTML document for the requested visualization (charts, diagrams, dashboards) and it renders inline in the chat inside a sandboxed iframe — supporting Tailwind CSS and external visualization libraries (Chart.js, Mermaid, D3, Plotly) loaded via CDN.

The in-product tool description reads: "Renders an interactive HTML visualization in the chat. Provide a description and data — the tool generates the HTML automatically using available libraries (Chart.js, Mermaid, D3, Plotly, etc.)."

Settings → Tools → Create Tool — Data Visualization appears in the Add Tool list alongside User Memory, Context Memory, Agent to Agent, and other built-in types Data Visualization tool configuration — Tool Name, Description, Visualization Instructions, Max Output Tokens (default 16000), Temperature (default 0.2) Sample Product Pricing Analysis rendered inline in chat — Account Manager card, Inventory Pricing Summary table with per-product original price, discount, final price and status, plus totals

Key Capabilities¶

• Built-in tool type — Data Visualization appears as a new entry in the Add Tool list alongside Agent to Agent, User Memory, Context Memory, Human Agent, Salesforce / HubSpot Create Lead, etc.
• Sandboxed rendering — Generated HTML renders inside an iframe in the chat, free to use external CDN libraries and Tailwind CSS without affecting the host page.
• Author-controlled defaults — Visualization Instructions are appended to the system prompt for the generation call, so admins can lock in brand colors, chart defaults, and styling preferences without re-stating them on every query.

Configuration Fields¶

How to Configure & Use¶

1. Create the tool — Navigate to Settings > Processors > Tools > Create Tool and select Data Visualization from the Add Tool list.
2. Configure parameters — Fill in Tool Name, Description, optional Visualization Instructions, and tune Max Output Tokens / Temperature if needed, then Save.
3. Add to an agent — Open the target agent's Settings > Tools section → Add, search for the tool by name, and select it from the picker.
4. Generate visualizations — The agent now renders visualizations inline in chat based on the user's prompt — e.g., a product pricing analysis card showing per-product original price, discount, final price and status, with totals.

Available Libraries¶

Rules for AI-Generated Visualizations¶

• Use only Tailwind CSS classes for styling (loaded automatically)
• Only request libraries actually used
• Write clean, self-contained HTML with inline <script> tags
• Content renders inside an iframe — no external navigation or form submissions
• Keep visualizations responsive and accessible

Assigning Tools to Agents¶

1. Navigate to Agent Settings > Tools
2. Click "Add Tool"
3. Select a configured tool from the list
4. Configure tool-specific settings for this agent
5. Click "Save"

Related Topics¶

• Pre-Processors — Process user input before the agent responds
• Post-Processors — Process conversations after responses
• Agent Builder — Advanced Configuration — Assign tools to agents
• Knowledge Overview — Set up knowledge repositories for Context and Retrieval tools
• Back to Settings