Subprocessors
Last updated: 4 September 2026
This is the authoritative list of third parties that process Speech customer data on our behalf. It is incorporated by reference into Annex 2 of our Data Processing Addendum. If you have an agreement with us, we will tell you before we add a new subprocessor that touches your content.
Processors that see what you dictate
| Company | Purpose | What it receives | Where |
|---|---|---|---|
| Sarvam AI | Speech to text | Your audio | India |
| OpenRouter | Model routing | Routes the request on our behalf to the model that runs it. Depending on the path a dictation takes that is Google, which may receive the audio itself, or Groq, which receives the transcribed text. | United States |
| OpenAI | Text commands and rewrites | The selected text you asked to change. Never audio. | United States |
In Privacy Mode these providers still transcribe the dictation, because that is the product. What changes is that nothing is retained afterwards: neither the audio nor the transcript is written to our database or our disks. See the privacy policy, section 6.
Processors that support the service without seeing what you dictate
| Company | Purpose | What it receives | Where |
|---|---|---|---|
| Amazon Web Services | Hosting, database, backup storage | Everything we store, as our infrastructure provider | India (ap-south-1, Mumbai) |
| Zoho (ZeptoMail) | Transactional email | Your email address and the sign-in code or service message being sent | India |
| Sign in with Google | Confirms your identity and gives us your email address. Separate from Google's role as a model provider above. | United States |
Only if you connect them yourself
Optional integrations are off unless you connect them, and disconnecting deletes the access token we hold. Today that is Swiggy, for Agent Mode ordering.
Where your data lives
Our servers, database and backups are in AWS ap-south-1 (Mumbai), India. We do not currently offer a region outside India. Two of the model providers above run in the United States, so a dictation that takes those paths crosses a border for the seconds it takes to process; nothing about it is stored there.
Back to the trust page · Privacy policy · Data Processing Addendum
Speech by Creativefuel