Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.nixflex.com/llms.txt

Use this file to discover all available pages before exploring further.

Custom functions extend the agent beyond built-in actions. Define a function (name, description, parameters, your URL), and Claude can call it during the conversation — to look up customer records, check inventory, book appointments, anything.

​
How it works

1

Define the function on the agent

Tell Nixflex the function’s name, what it does, and what parameters it takes.
2

Claude sees it as a tool

During calls, Claude is aware of available functions and decides when to call them.
3

Engine calls your URL

When Claude triggers a function, the engine makes an HTTP POST to your webhook with the parameters.
4

You return data

Your endpoint returns JSON. The engine feeds it back to Claude as context.
5

Claude responds

Claude uses your data to compose the next reply to the caller.

​
Defining a function

When you create or update an agent, include a functions array: 
{
  "name": "Acme Dental Receptionist",
  "system_prompt": "...",
  "functions": [
    {
      "name": "check_availability",
      "description": "Check if a specific date and time has availability for an appointment",
      "url": "https://api.acmedental.com/nixflex/availability",
      "method": "POST",
      "parameters": {
        "type": "object",
        "properties": {
          "date": {
            "type": "string",
            "description": "Date in YYYY-MM-DD format"
          },
          "time": {
            "type": "string",
            "description": "Time in 24-hour HH:MM format"
          }
        },
        "required": ["date", "time"]
      }
    }
  ]
}

​
What your endpoint receives

When Claude calls the function, your URL gets a POST request: 
{
  "function_name": "check_availability",
  "call_id": "CA1234567890",
  "caller_number": "+447386172392",
  "parameters": {
    "date": "2026-03-08",
    "time": "14:00"
  }
}
You respond with whatever structured data makes sense: 
{
  "available": true,
  "alternatives": ["14:30", "15:00", "15:30"]
}
The engine feeds your response to Claude, which composes the next reply:
“Yes, Tuesday the 8th at 2pm is free — would you like me to book that for you?”

​
Common use cases

Calendar lookups

Check availability before promising a slot.

CRM record fetches

Pull customer history when they call.

Booking creation

Create the booking in your system once confirmed.

Inventory / stock checks

Verify a product is available before quoting.

Order status

“Where’s my order?” — look it up and reply naturally.

Payment links

Generate a Stripe checkout URL and text it via [SEND_SMS:].

​
Timing constraints

Your endpoint must respond within 3 seconds. Anything slower and the caller experiences awkward silence. If your endpoint times out or errors:
  • The engine tells Claude “function failed”
  • Claude handles it gracefully (“I’m having trouble reaching the booking system, let me transfer you”)
  • The error is logged on the call record
For slow operations (sending an email, processing a refund), accept the function call and return immediately with { "queued": true }. Do the actual work in the background. Don’t block the call.

​
Authentication

Your endpoint should verify requests come from Nixflex. Each request includes a signature header: 
X-Nixflex-Signature: t=1715900000,v1=abc123...
Verify it with your webhook secret (set in the dashboard under your agent’s webhook config). See Webhooks for the verification algorithm.

​
Mentioning functions in the prompt

Tell Claude when to use each function: 
ACTIONS
- Before promising any appointment time, check availability first: use the check_availability function.
- Once the caller confirms a time, create the booking: use the create_booking function.
- For order status questions, look up the order: use the get_order function.
Claude will call the functions in the right order based on conversation flow.

​
Limitations

  • Functions run on each turn as needed; there’s no streaming partial results back to Claude
  • The engine doesn’t retry failed functions — your endpoint should be idempotent
  • The total context window is finite; large function responses are truncated
For more standardised integrations (Stripe, Salesforce, etc.), MCP server support is on the roadmap.