Create Tool

What is a Tool?

A Tool is a configurable module within Arahi AI that enables an Agent to perform specific actions—such as calling APIs, processing documents, or interacting with external systems. By building Tools, you extend your Agent’s capabilities from conversation only to real-world tasks.
Tools help your Agents act rather than just respond. :contentReference[oaicite:2]index=2

Steps to Create a Tool

  1. Navigate to the Tools section in your Arahi AI dashboard.

  2. Click Create Tool to begin.

  3. Enter a title/name for your Tool; it will appear in the metadata panel. :contentReference[oaicite:3]index=3

  4. Add metadata: tool icon, description, categories, tags (for governance and discoverability).

  5. Define the input schema: what inputs the Tool expects (fields, types, optional/required).

  6. Define the steps in the Tool’s logic: can include prompt steps, API calls, code execution, branching logic. :contentReference[oaicite:4]index=4

  7. Define the output schema: what outputs the Tool will return (fields, types).

  8. Configure advanced settings (optional): retry logic, timeout, authentication, connected service credentials.

  9. Save and test the Tool to ensure it behaves as expected.

Tool Builder Features

  • No-code / Low-code interface: Visual canvas to define tool steps, logic, and flows. :contentReference[oaicite:5]index=5

  • Pre-built templates: Use starter templates for common tool types (API call, document parser, notifier).

  • Custom code support: For advanced workflows you can embed custom code or run API steps. :contentReference[oaicite:6]index=6

  • Input variable mapping: Use placeholders for dynamic inputs that Agents will supply at runtime.

  • Output chaining: A Tool’s output can feed into another Tool or Agent prompt for workflow chaining.

Best Practices

  • Name your Tools clearly and concisely so other team members know their purpose.

  • Limit each Tool to one well-defined responsibility (single-responsibility principle).

  • Document the input/output schema and dependencies (e.g., required credentials).

  • Test Tools with edge-cases and failure scenarios (e.g., missing data, API errors).

  • Monitor tool execution logs and performance once live to catch inefficiencies or failures.

Example Tool: “Invoice PDF Parser”

  • Title: Invoice PDF Parser

  • Description: Extracts supplier, date, line items, total from uploaded invoice PDFs.

  • Input Schema:

    • invoice_pdf_url (string, required)

    • company_id (string, optional)

  • Step 1 (Prompt): “You are a PDF parsing assistant…” → extract raw text.

  • Step 2 (API or code): Run OCR or document-parser API on the PDF.

  • Step 3 (Prompt): Convert extracted text into structured JSON (supplier, date, items).

  • Output Schema:

    • supplier_name (string)

    • invoice_date (date)

    • line_items (array of objects: description, quantity, price)

    • total_amount (number)

  • After building, test with sample PDFs and validate output before connecting it to an Agent.

Updated on
Introduction
Delete a Tool