Skip to main content
After every call, Nixflex runs an analysis pass over the transcript. You get a summary, the caller’s sentiment, whether the call succeeded, and any custom fields you’ve defined — automatically. No setup required for the defaults.

​
What you get out of the box

Every call record has these fields populated automatically:
FieldTypeDescription
call_summarystring1–2 sentence summary of what happened
caller_sentimentenumhappy, neutral, or frustrated
call_successfulbooleanDid the caller get what they needed?
extracted_dataobjectYour custom fields (see below). Empty {} if none configured.
These are the field names on the call record returned by GET /v1/calls/:id and GET /v1/calls. In the webhook payload the same data is delivered under flat, renamed keys: summary, sentiment, successful, and extracted.

​
Custom data extraction

Beyond the defaults, you define exactly which fields to pull from each call. Set data_extraction_fields on the agent — an array of field definitions: 
{
  "data_extraction_fields": [
    {
      "type": "text",
      "name": "caller_name",
      "description": "The caller's first name",
      "optional": true
    },
    {
      "type": "boolean",
      "name": "wants_callback",
      "description": "Whether the caller asked for a callback"
    },
    {
      "type": "selector",
      "name": "service_requested",
      "description": "Which service the caller asked about",
      "choices": ["check-up", "cleaning", "whitening"]
    },
    {
      "type": "number",
      "name": "party_size",
      "description": "Number of people for a booking"
    }
  ]
}
For each call, the agent reads the transcript and fills in what it can. Missing fields come back as null — the agent won’t invent data.

​
Field types

TypeReturnsUse for
textstringNames, emails, free-form values. Optional format_example guides the format.
selectorstring from choicesCategorising into a fixed set of options. Requires choices.
booleantrue / falseYes/no questions (callback wanted, issue resolved).
numbernumberQuantities, counts, party sizes.

​
Field properties

PropertyRequiredNotes
typeyesOne of text, selector, boolean, number
nameyesKey used in extracted_data. Keep it snake_case.
descriptionyesWhat the agent should look for. Be specific — this drives accuracy.
optionalnoIf true, the field stays null when not mentioned rather than guessed
choicesselector onlyThe allowed values
format_exampletext onlyExample of the expected format, e.g. name@example.com
The result lands in extracted_data: 
{
  "extracted_data": {
    "caller_name": "Amir",
    "wants_callback": true,
    "service_requested": "check-up",
    "party_size": null
  }
}
On the API, your custom fields are keyed under extracted_data. In the webhook payload the same fields arrive under extracted.

​
When analysis runs

  • Inbound, outbound, and batch calls — all directions, always.
  • Calls with real conversation — if the caller spoke fewer than 2 turns or fewer than 10 words total (hang-ups, voicemail), the AI pass is skipped to save cost. Defaults fall back to a short summary and call_successful: false.
  • Asynchronously, after the call ends — by the time your webhook fires (~60s later), all fields are populated.

​
How extraction works under the hood

Analysis runs in two stages:
  1. Base analysis — always runs. One AI call returns call_summary, caller_sentiment, call_successful.
  2. Custom extraction — only runs if the agent has data_extraction_fields. A second AI call returns your custom fields.
Both stages use the AI and add roughly $0.001 to the call cost. It’s included in the per-minute price — no extra charge.

​
Configuring in the dashboard

Open your agent’s Post-Call Data Extraction panel. Add fields with the + Add button — pick a type, give it a name and description, and save. Extracted fields are delivered to your webhook after each call. No code required.

​
Reading analysis programmatically

Each call object includes the analysis inline — there’s no separate endpoint. See Get call for the full schema, or receive it via webhook.