Docs

Getting Started

Install, configure env vars, and launch Talkform locally or on Vercel.

Talkform turns a structured form into a live audio interview.

Install

pnpm install
pnpm dev

The hosted app runs from apps/web.

Required environment variables

  • OPENAI_API_KEY
  • OPENAI_REALTIME_MODEL optional, defaults to gpt-realtime-2.1
  • OPENAI_REALTIME_VOICE optional, defaults to marin
  • TALKFORM_ENABLE_IMPORT_AI_REFINEMENT optional; defaults off so imports remain deterministic and do not call a paid copy-refinement model
  • NEXT_PUBLIC_AUDIOFORM_VENDOR optional
  • NEXT_PUBLIC_AUDIOFORM_VENDOR_URL optional
  • TALKFORM_API_TOKEN optional server-only token for authenticated CLI or another trusted machine client
  • AUDIOFORM_API_TOKEN matching token in the CLI process, not in browser code

Local demo

  1. Start the app with pnpm dev
  2. Open /app
  3. Allow microphone access
  4. Click Start onboarding

Vercel

Transient session APIs and public Realtime client-secret issuance are disabled in hosted production by default. A production deployment needs a durable session store, a distributed rate limiter, and server authentication before those endpoints should be enabled. The checked-in session store and quotas are process-local reference implementations.

Link the app to your Vercel project and define env vars for development, preview, and production:

vercel link
vercel env add OPENAI_API_KEY production
vercel env add OPENAI_API_KEY preview
vercel env add OPENAI_API_KEY development

For authenticated machine access, add TALKFORM_API_TOKEN to the server environment and set the same value as AUDIOFORM_API_TOKEN only in the trusted CLI process. Never create a NEXT_PUBLIC_ token.

Controlled, non-public deployments can opt into the reference routes with TALKFORM_ENABLE_IN_MEMORY_SESSIONS=true and TALKFORM_ENABLE_PUBLIC_REALTIME=true. Do not set these on hosted production until the process-local store and quota maps have been replaced by durable, distributed services.

Attach the production domain talkform.ai to the Vercel project once the first deployment exists.