How tools extend your AI assistant

Tools Overview

Tools let the AI call out to external data sources or perform actions during a conversation. They transform your AI from a simple chatbot into a capable agent that can solve real-world problems.

Data Acquisition vs. Action Tools

Vindex Ai supports two main categories of tools:

1. Data Acquisition (Read-Only)

These tools fetch real-time information to help the AI answer questions accurately.

Example: Fetching product pricing, checking order status, or searching a knowledge base.
Methods: Usually
text
```
GET
```
or
text
```
POST
```
(for search).
Caching: Supported via
text
```
cache_ttl
```
to improve performance.

2. Action-Oriented Tools (Read-Write)

These tools perform tasks in external systems.

Example: Creating a support ticket, subscribing a user to a newsletter, or updating a shipping address.
Methods:
text
```
POST
```
,
text
```
PUT
```
,
text
```
PATCH
```
, or
text
```
DELETE
```
.
Confirmation: Can be configured to require explicit user confirmation before executing.

Human-in-the-Loop Confirmation

For safety, any tool can be configured with Requires Confirmation. When enabled:

The AI identifies that it needs to use the tool.
The AI pauses and presents the details of the action to the user (e.g., "I'm about to create a ticket with the subject 'Login Issue', should I proceed?").
The action is only executed after the user gives explicit consent (e.g., clicking "Confirm" or saying "Yes").

Custom API Tools

You can define custom tools for any Channel. Each tool supports:

Protocol: REST, GraphQL, or Form POST.
Dynamic Parameters: Define variables that the AI will extract from the conversation.
Response Mapping: Use
text
```
response_path
```
(dot notation) to extract exactly what the AI needs from complex JSON responses.
Security: Custom headers for API keys and authentication.

Authenticated & Internal Tools

One of the most powerful uses of the Vindex Ai Gateway is connecting agents to internal, authenticated data systems like School Management Systems, Hospital Portals, or Internal CRMs.

Secure Communication via Headers

By using custom Headers, you can authorize the Gateway to speak to your internal APIs securely.

How to set up Tool Authentication:

Prepare your API: Ensure your internal system has an API endpoint that accepts an authentication token (e.g., a Bearer token or API Key).
Configure the Tool: In the Vindex Ai Dashboard, navigate to your Channel's API Tools and click New Tool.
Add Auth Headers: Scroll to the Headers section and add your credentials:
- Key:
  text
```
Authorization
```
- Value:
  text
```
Bearer your_secret_service_token
```
Secrets remain Secret: These headers are stored encrypted on our backend. The end-user chatting with the AI never sees these credentials; they are injected by the Gateway only during the outgoing request to your API.

Safe Data Management

When managing internal records (e.g., updating a student's grade or scheduling a patient), use the Requires Confirmation flag. This ensures the AI acts only as a "proposer" while a human maintains ultimate control.

Example: School Management Interface

Read Action: A teacher asks, "Who is John Doe's guardian?" The agent calls
text
```
get_student_info
```
using a
text
```
GET
```
request with your
text
```
Authorization
```
header. It retrieves the data and answers immediately.
Write Action: The teacher says, "Update attendance: John Doe is Present." The agent prepares the
text
```
update_attendance
```
call. Because this is a sensitive "Write" action, the Gateway halts the request and asks the teacher for a final confirmation before the record is changed.

Built-in Tools

Web Scraper: Automatically enabled when a Website URL is configured. Allows the AI to "read" your site.
Knowledge Search: Powered by vector embeddings of your uploaded documents.
Contact Tool: A specialized tool for capturing leads and sending them to your email.

Limits & Best Practices

Response Limits: Keep tool outputs under 3,000 characters to avoid blowing the AI's token budget.
Active Tools: Each Channel is limited to 5 active custom tools to maintain optimal performance and accuracy.
Descriptive Names: Give your tools clear, lowercase snake_case names (e.g.,
text
```
create_support_ticket
```
) so the AI knows exactly when to use them.