Settings

Organization settings, the analysis prompts, the extraction fields, and model routing. The last two have the widest blast radius — they change how every call is analysed.

Organization settings

GET /api/settings

{
  "report_recipients": ["[email protected]"],
  "call_context": "Outbound calls selling ergonomic office furniture to SMBs.",
  "stt_model": "gpu",
  "auto_reports_enabled": true,
  "is_premium": false,
  "dismissed_default_scores": ["small_talk"]
}

PUT /api/settings — manager or above

{
  "report_recipients": ["[email protected]", "[email protected]"],
  "call_context": "Outbound calls selling ergonomic office furniture to SMBs.",
  "auto_reports_enabled": true
}

Report recipients

Who receives generated reports. Plain addresses; they do not need to be members of the organization.

Dismissed default scores

PATCH /api/settings/dismissed-default-scores

{ "dismissed_default_scores": ["small_talk", "call_opening"] }

Hides built-in scoring criteria that do not apply to your calls. The score is still computed — this only removes it from display, so dismissing one and changing your mind later costs nothing.

---

Analysis prompts

GET /api/settings/analysis-prompts

{
  "system_prompt": "You are analysing B2B furniture sales calls…",
  "safety_prompt": "Never infer a customer's health, religion, or finances…",
  "default_system_prompt": "…",
  "default_safety_prompt": "…",
  "system_prompt_uses_default": false,
  "safety_prompt_uses_default": true,
  "combined_prompt_preview": "…"
}

The defaults are returned alongside your overrides, so a UI can show what you are diverging from — and combined_prompt_preview shows exactly what the model will receive, which is the only reliable way to check a prompt edit did what you meant.

PATCH /api/settings/analysis-prompts — manager or above

{
  "system_prompt": "You are analysing B2B furniture sales calls…",
  "set_system_prompt": true,
  "safety_prompt": null,
  "set_safety_prompt": false
}

The set_* booleans distinguish "set this to the value I sent" from "leave it alone" — which is what lets you clear a prompt back to the default by sending null with set_*: true, rather than an omitted field meaning two things.

Prompt changes do not re-analyse existing calls

They apply to calls analysed afterwards. Use POST /api/calls/{id}/reanalyze to bring an existing call up to date — superadmin only, because doing it across a deployment is expensive.

---

Extraction fields

The structured questions asked of every transcript. Reading them is open to any member; changing them is superadmin only, because they apply across the whole deployment.

List fields

GET /api/extraction-fields — superadmin

{
  "fields": [
    {
      "id": 7,
      "uuid": "ef1a2b3c-…",
      "name": "budget_confirmed",
      "field_type": "boolean",
      "description": "Did the customer confirm they have budget approved?",
      "options": [],
      "min_value": null,
      "max_value": null,
      "scope": ["outbound"],
      "category": "qualification",
      "enabled": true
    }
  ]
}

Create a field

POST /api/extraction-fields — superadmin

{
  "name": "budget_confirmed",
  "field_type": "boolean",
  "description": "Did the customer confirm they have budget approved?",
  "scope": ["outbound"],
  "options": [],
  "min_value": null,
  "max_value": null,
  "category": "qualification",
  "enabled": true
}

Field	Required	Notes
name	✓	The key this lands under in a call's custom_field_values
field_type	✓	boolean, string, number, enum
scope	✓	Restricts the field to inbound, outbound, or both
description	—	What the model is told to look for — write it as an instruction
options	—	Allowed values for enum
min_value / max_value	—	Bounds for number
category	—	Grouping for display
enabled	—	Disable without deleting, preserving historical values

Update a field

PATCH /api/extraction-fields/{id} — superadmin

Every field above is optional on update, plus one that only exists here:

Field	Notes
prompt_override	Replaces the generated instruction for this field with your own wording

Delete a field

DELETE /api/extraction-fields/{id} — superadmin

Field labels

GET /api/extraction-field-labels

Available to any member, and to API keys with sales:analyses:read. Returns just the UUID, name, and type of each field — enough to render a call's custom_field_values without exposing the prompts behind them.

description is a prompt, not a label

"Did the customer confirm they have budget approved?" extracts reliably. "Budget" does not. If a field is coming back empty or wrong across many calls, the description is usually why.

---

Model routing

GET /api/settings/model-routing

{
  "stt_model": "gpu",
  "diarization_url": "https://…",
  "translation": { "enabled": true, "provider": "openai", "vllm_endpoint": null },
  "custom_analysis": { "enabled": true, "provider": "vllm", "vllm_endpoint": "http://localhost:8000/v1" }
}

PATCH /api/settings/model-routing — superadmin only

stt_model accepts gpu or sarvam. The remaining keys choose which model serves each stage: diarization, translation, and the custom analysis pass. provider: "vllm" points a stage at self-hosted inference through vllm_endpoint, which is how an on-prem deployment keeps transcripts inside its own network.

This is superadmin-gated in every binary that has it — a misrouted model breaks analysis for every organization on the deployment, not just the one that changed it.

---

Call layout

Control how a call's analysis is arranged in the console. Display only — nothing here changes what is extracted.

List sections

GET /api/call-sections — any member

{
  "sections": [
    { "key": "__section:scores", "label": "Scores" },
    { "key": "__section:key_details", "label": "Key Details" }
  ]
}

Set the layout

PUT /api/call-layout — superadmin

{
  "hidden": ["waveform"],
  "collapsed": ["key_details"]
}

Field	Notes
hidden	Section ids not shown at all
collapsed	Section ids shown collapsed by default

Order the signal groups

PUT /api/signal-group-order — superadmin

{ "group_order": ["qualification", "outcome", "next_action"] }