Configuration

Talkform is driven by AudioformConfig.

Shape

type AudioformConfig = {
  id: string;
  title: string;
  description?: string;
  instructions?: string;
  fields: AudioformField[];
  mode?: "text" | "voice";
  branding?: {
    fromName?: string;
    purpose?: string;
    logoUrl?: string;
    wordmark?: string;
    faviconUrl?: string;
    fontFamily?: string;
    showPoweredBy?: boolean;
  };
  theme?: {
    accent?: string;
    surface?: string;
    panel?: string;
  };
  realtime?: {
    model?: string;
    voice?: string;
  };
  output?: {
    formats?: Array<"json" | "markdown">;
    webhookUrl?: string;
  };
};

Hosted respondent pages are product-owned by default: no Talkform marketing chrome. Set branding.fromName and branding.purpose so the first screen answers who the interview is from and what it is for. Theme colors must be hex values such as #1c1917. showPoweredBy stays off unless you set it to true. promptTitle is the question the person reads or hears; label is the field name in the review panel.

Set mode: "text" when the interview should not offer a microphone. Voice is only available on a claimed human-owned project.

The output.webhookUrl field remains part of the general config schema for local and legacy workflows. Hosted handoff delivery uses the project-scoped PUT /api/v1/webhook endpoint so the endpoint, signing secret, retries, and delivery status are managed separately from each form config.

Field types

• text
• long_text
• single_select
• multi_select
• number
• rating
• url
• file_ref

Every field includes:

• id
• label
• type
• required
• promptTitle
• promptDetail
• options?
• validation?
• agentHint?

Design guidance

• Keep prompts short and concrete.
• Use single_select and multi_select whenever the downstream system expects enumerated values.
• Put product-specific logic downstream of the exported JSON instead of inside the Talkform core.