Understanding Tools

Tools Enable AI Actions Tools are the bridge between the AI agent and your Salesforce org. They let the agent actively retrieve, query, and modify data during conversations.

What are Tools?

Tools in REDA AI are pre-built integrations that allow the AI agent to interact with Salesforce. Without tools, the AI can only provide suggestions based on prompts and static context. With tools, the AI can actively query records, create new data, update fields, and perform actions.

Tools are defined as Custom Metadata records under the hood. REDA AI ships with a comprehensive set of built-in tools that cover common Salesforce operations, which you assign to your agents as needed. When a built-in tool doesn't cover what you need, you can also build your own custom tool — see Configuring Tools.

Two Categories of Tools

Category	Field Value	Capability	Examples
Tool (Read-Only)	Tool	Query Salesforce data without changing anything	soql_query, get_record, count_records, describe_object, get_picklist_values, get_record_types
Action (Read-Write)	Action	Create, update, delete records or trigger actions	create_record, update_record, delete_record, send_email, create_task, create_event, schedule_action, link_file_to_record, send_custom_notification

Common Out-of-the-Box Tools

Tool Name	Category	Purpose
soql_query	Tool	Query records using SOQL with filters, sorting, and field selection
get_record	Tool	Retrieve a single record by ID and return all accessible fields
count_records	Tool	Count records matching criteria (e.g. “how many open cases?”)
describe_object	Tool	Get metadata about a Salesforce object (field names, types, relationships)
get_picklist_values	Tool	Get the picklist values for a specific field
get_record_types	Tool	Get available record types for an object
create_record	Action	Create a new Salesforce record with specified field values
update_record	Action	Update specific fields on an existing record
delete_record	Action	Delete a record (moved to recycle bin)
send_email	Action	Send an email from the user to specified recipients
create_task	Action	Create a new task record
create_event	Action	Create a calendar event
schedule_action	Action	Schedule a future action (executes without confirmation)
link_file_to_record	Action	Attach an uploaded file to a Salesforce record
send_custom_notification	Action	Send a Salesforce custom notification to specified users

Additional Capabilities

Beyond the tools listed above, REDA AI agents can also use Web Search (search the internet for real-time information) and File Upload (process uploaded documents, images, and CSV data). These are enabled per agent on the Agent Definition — see their dedicated pages for details.

How the AI Decides to Call Tools

When a user sends a message, the AI processes it as follows:

AI Receives Message
The user asks a question or requests an action.
AI Analyzes Available Tools
The AI reviews the list of tools assigned to this agent.
AI Decides If a Tool is Needed
Based on the user's question, the AI decides if calling a tool would help. Example: "To answer this question, I need to query the Opportunities for this account" → calls soql_query.
AI Constructs the Tool Call
The AI formulates the parameters needed. Example: soql_query(object="Opportunity", filters={"AccountId": "001xx000003DHP"})
REDA AI Executes the Tool
REDA AI intercepts the tool call, validates permissions, and executes it against Salesforce.
Results Returned to AI
The tool result is sent back to the AI. Example: [{"Id": "006xx000001SMeAAM", "StageName": "Negotiation/Review", "Amount": 50000}, ...]
AI Formulates Response
The AI uses the tool result to answer the user's original question with specific data.

Action Tools Require Confirmation

On the Internal channel, before executing an Action tool whose Requires Confirmation is on (create_record, send_email, update_record and four others), the agent shows a confirmation card. The confirmation card displays:

What action will be performed (e.g., "Create Task")
Key field values that will be set (e.g., Subject, DueDate, Priority)
A "Confirm" button to proceed or "Cancel" button to abort

This prevents accidental data changes. The user must consciously approve each action before it executes.

Two limits on that. Three Action tools have Requires Confirmation off and never show a card — schedule_action, link_file_to_record and continue_workflow. And cards exist only on the Internal channel: on External Widget and Community the agent executes the action and narrates the outcome instead, and on the API channel a Flow decides via Auto Execute Actions. See Action Confirmations.

REDA AI respects Salesforce's security model. When the AI calls a tool:

Field-Level Security (FLS) — The AI can only access fields that the running user has read or write access to. Fields the user can't access are excluded.
Object-Level Permissions — The AI can only query or create records on objects the user has access to.
Record Sharing — The AI can only access records shared to the user via sharing rules, org-wide defaults, or role hierarchy.
One deliberate exception — an agent whose Execution Context is System and whose Channel is API reads fields the running user cannot. Even then, record sharing is never widened: it sees more fields on records it could already reach. Everywhere else REDA AI operates strictly as the logged-in user. See Data Security.

This means an AI agent is as secure as the user running it. A support agent accessed by a customer portal user can only see and modify records shared to that portal user.

Built-in Tools are Package-Managed

The built-in tools listed above are defined as Custom Metadata records managed by the REDA AI package. This means:

You cannot edit or remove a built-in tool's definition — only assign it to agents or leave it unassigned
The set of built-in tools is updated whenever the REDA AI package is upgraded
Built-in tool definitions are read-only in your org
You configure which tools to use by assigning them to agents

When a built-in tool doesn't cover what you need, build a custom tool instead — a SOQL Template, Apex Class, or File Mapping tool — in the Tool Builder. See Configuring Tools for the full walkthrough.

Tool Assignment to Agents

You don't use all tools on all agents. Instead:

Open the Agent Definition
Find the "Tools" related list
Click "Add Tool"
Select only the tools this agent needs
Save

Only assigned tools are available to the agent. Unassigned tools are never suggested by the AI.

Token Cost of Tools

Each tool available to the agent consumes tokens during each request. Even if the tool isn't called, its definition is included in the context sent to the AI. This means:

Assign only tools the agent actually needs
Remove unused tools to save tokens and reduce cost
A platform setting controls when tools are loaded — always, or only on record pages

For example, removing unnecessary tools from an agent noticeably reduces token consumption per request, directly lowering your credit costs. See Tool Loading & Cost for how the stable, provider-cached tool list keeps the remaining cost down.

Frequently Asked Questions

Yes. The built-in tools shipped by the REDA AI package are read-only, but you can build your own SOQL Template, Apex Class, or File Mapping tool in the Tool Builder whenever the built-ins don't cover your use case. See Configuring Tools for the full walkthrough.

The tool call fails with a permission error. The AI receives the error and typically informs the user "You don't have permission to access this data" or similar. The error doesn't break the conversation — the AI can suggest alternatives or ask for more information.

No. schedule_action is unique in that it executes silently without a confirmation card. This is because it schedules a future action rather than immediately modifying data, and future scheduled actions can be cancelled by the user afterward if needed.

Think about what the agent needs to do. If it's a Sales agent, it probably needs soql_query, create_task, create_event, send_email. If it's a read-only reference agent, you might only assign soql_query and describe_object. Use "Build Agent with AI" to get recommendations.

Next Steps

Learn how to configure tools for your agents, or explore action confirmation cards to understand how users approve data changes.

Next: Configuring Tools →