Sub-processors
Last updated 20 November 2026
The providers that help us run TalkDemo, what data each sees and where.
A sub-processor is a company that processes personal data on TalkDemo's behalf so we can provide the service. Each customer that uses TalkDemo is the data fiduciary (under India's DPDP Act) and controller (under the GDPR) for its prospects' data, and TalkDemo processes that data for the customer (see the data processing agreement). The providers below process it only for the purposes listed and under contract with us.
TalkDemo is pre-launch, in private pilots, so this list is planned and may change before launch. Regions marked “set at launch” are confirmed before any customer's data is processed there, and we update this page before any change, never after.
01Planned sub-processors
- Vercel
- What for:
- Hosting for the website, dashboard and call page
- Data:
- Requests (IP address, browser) and the data the app handles while it runs
- Where:
- Global edge; functions in Mumbai (bom1) for India
- Managed PostgreSQL (AWS RDS or Supabase, Mumbai region)
- What for:
- Database for account data and, separately, the logs
- Data:
- Accounts, agents, flows, sessions, transcripts, leads; audit and AI call logs
- Where:
- India (ap-south-1)
- Amazon Web Services
- What for:
- Private file storage, malware scanning and email
- Data:
- Recordings, screenshots, decks, knowledge files; email addresses and content
- Where:
- India (ap-south-1)
- LiveKit
- What for:
- Real-time calls: audio, screen share and recording
- Data:
- Call audio and the shared screen while the call runs
- Where:
- Region pinned to the prospect (Mumbai for India)
- Sarvam AI
- What for:
- Speech-to-text and text-to-speech for Indian languages, Hinglish and English (India)
- Data:
- Call audio and the agent's spoken text
- Where:
- India
- Anthropic (Claude)
- What for:
- The agent's conversation and reasoning, summaries and lead scoring
- Data:
- Transcript, your knowledge passages and screen summaries. Never demo credentials
- Where:
- United States; zero-retention settings where offered
- Browserbase
- What for:
- Isolated cloud browsers for live product demos (only for: agents in live product mode)
- Data:
- Your demo tenant's pages during a session
- Where:
- Nearest available region; Singapore for Indian sessions (no India region yet). It renders your demo tenant, not prospect data
- Deepgram
- What for:
- Speech-to-text for English (US, UK and Australia) and world languages (only for: only for calls in those languages, never for Indian languages)
- Data:
- Call audio
- Where:
- United States
- ElevenLabs
- What for:
- Text-to-speech for English (US, UK and Australia) and world languages (only for: only for calls in those languages, never for Indian languages)
- Data:
- The agent's spoken text
- Where:
- United States
- Cartesia
- What for:
- Backup text-to-speech for world languages (only for: only for world-language calls, if ElevenLabs is unavailable)
- Data:
- The agent's spoken text
- Where:
- United States
- OpenAI
- What for:
- Embeddings for knowledge search (text-embedding-3-small), and backup text-to-speech for world languages
- Data:
- Text of your knowledge sources (no prospect data); on a backup world-language voice, the agent's spoken text
- Where:
- United States
- Cloudflare
- What for:
- DNS and Turnstile bot checks on public forms
- Data:
- Browser signals during form checks
- Where:
- Global network
- Razorpay
- What for:
- Payments in rupees (UPI AutoPay, cards, GST invoices) (only for: customers paying in rupees)
- Data:
- Billing contact, invoices and payment status. Card and UPI details stay with Razorpay
- Where:
- India
- Stripe
- What for:
- Card payments in US dollars, with receipts and invoices (only for: customers paying in US dollars)
- Data:
- Billing contact, invoices and payment status. Card details stay with Stripe
- Where:
- United States
- cron-job.org
- What for:
- Runs scheduled jobs and uptime checks
- Data:
- No personal data: it calls TalkDemo's job addresses on a schedule
- Where:
- Germany
- Sentry, PostHog and Langfuse
- What for:
- Error tracking, product analytics and AI traces
- Data:
- Error details and usage events with personal data removed
- Where:
- EU or US (set at launch)
Providers marked “only for” process data only when a customer turns that feature on or pays that way. A provider is added here before it is switched on, never after.
02What stays with TalkDemo
- Demo credentials are never sent to any AI or speech provider. They are opened only by our browser service to log into the customer's demo tenant.
- The AI model provider receives the conversation, relevant passages from the customer's knowledge sources and screen summaries, only for the demo in progress. Another customer's data is never sent.
- AI providers are used under terms that don't train on submitted content, with zero-retention settings where offered. We confirm those terms for each provider before launch.
- Embedding providers receive only the text of the customer's knowledge sources, never prospect data.
- Logs never contain demo credentials, full contact details, raw IP addresses, passwords or tokens.
- Card, UPI and bank details stay with Razorpay and Stripe.
- TalkDemo doesn't sell personal data or share it for advertising.
03How changes are announced
Before a new sub-processor starts processing customer or prospect personal data, we add it to this page, change the date at the top, and email the Admins of every customer using TalkDemo. A customer that objects can tell us at hello@talkdemo.com, and we will work with it on a solution, as set out in the data processing agreement.
Removing a provider, or a change that doesn't affect personal data, is updated here without an email.
04Where data is hosted
For customers in India, the database and file storage run in AWS Mumbai (ap-south-1) and Indian-language speech runs on Sarvam in India. Real-time calls and cloud browsers are pinned to the region nearest the prospect. The AI model provider (Anthropic), the speech providers for world-language calls (Deepgram and ElevenLabs, never used for Indian languages) and the embedding provider process data in the United States, under EU standard contractual clauses and the UK transfer agreement where the GDPR or UK GDPR applies. More detail: the DPDP Act page, the GDPR page and the security page.