Skip to main content
A tool is something the agent can invoke mid-conversation: query your CRM, schedule a meeting, end a phone call, transfer to a human, or post to Slack. Agntix supports two types:
TypeWhat it is
API ToolAn HTTP endpoint Agntix calls on the agent’s behalf. You define URL, method, headers, query/body schema.
Function ToolA built-in capability — endPhoneCall, transferPhoneCall, detectVoicemail, and similar — that runs in our infrastructure.
This guide walks you through creating an API tool, attaching it to an agent, and testing the round-trip — all from the dashboard.

​
Prerequisites

  • A signed-in dashboard account at app.agntix.ai.
  • An existing agent (see Build your first agent).
  • For API tools: a reachable HTTPS endpoint you want the agent to call. The endpoint should respond in < 1 second for voice agents to stay snappy.

​
Step 1: Open the Tools page

From the left navigation, click Tools. New organizations start with an empty list.
Empty tools list
Click + Create Tool.

​
Step 2: Fill in the basics

The Create Tool modal collects:
  1. Name — 8–20 chars, lowercase letters, digits, and underscores only (e.g. lookup_order_status). The LLM sees this name verbatim — make it self-describing.
  2. Description — what the tool does. The LLM uses this to decide when to call it — be precise. Bad: “queries orders”. Good: “Looks up the current status of a customer order by its order number. Returns status, expected delivery date, and shipping carrier.”
  3. Type — leave on API Tool for this guide.
Create Tool modal

​
Step 3: Configure the HTTP request

Below the basics you’ll see the request configuration:
  • URL — full HTTPS URL. May contain {placeholders} that Agntix fills from the schema you define in Step 4 (e.g. https://api.example.com/orders/{orderId}).
  • Method — GET, POST, PUT, PATCH, DELETE.
  • Body — JSON template (for non-GET methods). Placeholders work here too.
  • Headers — key/value pairs. Common pattern: Authorization: Bearer {YOUR_TOKEN}.
  • Query Params — key/value pairs appended to the URL.
Headers and params
Already have a working curl command? Click the Import from curl button at the top of the modal — Agntix parses it and fills URL, method, headers, body, and params for you.

​
Step 4: Define the input schema

The LLM needs to know which arguments to pass when calling the tool. Click Schema Builder to open the visual editor, or paste a JSON Schema directly. For the order-status example: 
{
  "type": "object",
  "properties": {
    "orderId": {
      "type": "string",
      "description": "The order number to look up, e.g. 'ORD-1234'."
    }
  },
  "required": ["orderId"]
}
Schema builder
Click Save to return to the tool modal. Your placeholders ({orderId} in the URL) now map to schema properties.

​
Step 5: Test the API

Before saving, click Test API in the modal. A side sheet opens where you can fill in sample values for each schema property and fire a real request against your endpoint.
API test success
Tweak headers / URL / schema until you get a 2xx response. Tool calls share the voice/chat latency budget — anything slower than ~1 s will noticeably hurt voice UX. Close the side sheet and click Create Tool in the main modal.

​
Step 6: Attach the tool to an agent

Open your agent at /agents/{id}. Switch to the Functions tab on the right side. The tab shows two groups:
  • Built-in functions — toggle on for End Call, Transfer Call, Voicemail detection. These are pre-configured; you only pick when they fire.
  • Custom Tools — click + Add to attach your newly-created tool. You can attach multiple.
Attach tool to agent
Click Save Changes in the page header.

​
Step 7: Verify the tool fires

Click Test Agent in the page header. Send a message that should obviously trigger the tool — for the order-status example:
“What’s the status of order ORD-1234?”
In the test panel you’ll see three things in sequence:
  1. The user message you sent.
  2. A tool call entry — lookup_order_status invoked with {"orderId": "ORD-1234"}.
  3. The assistant’s final reply, incorporating the tool’s response.
Tool fired in test
You can also see this in Logs / History for any real session — the tool call shows up as a TOOL message with full payload.

​
Verify it works

  • The tool appears in the /tools list with the correct name and type.
  • The agent’s Functions tab shows the tool selected after saving.
  • A test message that should trigger the tool actually fires the HTTP request (check your server logs).
  • The agent’s final reply uses information from the tool’s response.

​
Troubleshooting

SymptomLikely causeFix
Tool never firesDescription too vagueRewrite the Description to explicitly say when the tool should be used; restate examples
Tool fires too oftenDescription too aggressiveSoften the description: “Use only when the user provides an order number.”
401/403 from your APIHeader placeholder not filledDouble-check the Headers section; ensure no literal {YOUR_TOKEN} text
Voice latency spike when tool firesSlow endpointCache server-side; aim for < 500 ms p95
Agent invents tool argumentsRequired fields missing in schemaMark fields required in the JSON Schema

​
Function tools

For end call / transfer call / voicemail detection, you don’t need to build an API. Open the agent’s Functions tab and toggle the built-in function on — Agntix handles the implementation. Configure conditions (e.g. transfer destination phone number) in the same panel.

​
API alternative

Same flow over HTTP: 
curl -X POST https://api.agntix.ai/v1/chat/tools \
  -H "x-api-key: $AGNTIX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "lookup_order_status",
    "description": "Look up the current status of an order by its number.",
    "type": "API",
    "apiTool": {
      "url": "https://api.example.com/orders/{orderId}",
      "method": "GET",
      "headers": { "Authorization": "Bearer $YOUR_API_TOKEN" }
    },
    "ajvPropertiesSchema": {
      "type": "object",
      "properties": { "orderId": { "type": "string" } },
      "required": ["orderId"]
    }
  }'

curl -X PATCH https://api.agntix.ai/v1/chat/agents/$AGENT_ID \
  -H "x-api-key: $AGNTIX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "config": { "toolsEnabled": true, "toolIds": ["tool_…"] } }'
See the Tools API reference for the full schema.

​
Common next steps

Streaming events

Watch tool calls live from your backend.

Voice calls

Tools work the same way over voice — with tighter latency budgets.

Webhooks

Get notified server-side when a tool fires.

Knowledge stores

Pair tools with RAG for the strongest agents.