Loading...
REDENVOY
Where your customers' voices go, who holds the keys, and which rules the product enforces. Stated plainly, including what is not finished.
RedEnvoy is taking its first pilot clients. Everything marked In the product below is built and tested, but it has not yet run a paid client campaign. We will say the same in your security review.
Who holds what
The same codebase runs either way. Managed is fastest to start. Self-hosted keeps everything inside your own cloud account, with your own keys.
| Managed by Red Augment | Self-hosted in your cloud | |
|---|---|---|
| Where it runs | Red Augment's AWS account, Mumbai | Your cloud account, single tenant |
| Speech to text, text to speech | Sarvam and other speech providers, agreed per campaign | The same providers, on your accounts |
| Conversation model | OpenAI, United States | OpenAI on your account, or a model we scope with you |
| Telephony | Plivo, on your registered number | Your SIP trunk and your numbers |
| Recordings & transcripts | Stored by us in Mumbai. Audio deleted at 90 days. | Your storage, your retention |
| Provider keys | Held by us per deployment, encrypted | Yours. Red Augment never sees them. |
| Results file | Delivered to you | Written in your environment |
Call audio is processed in India. The text of the conversation goes to OpenAI in the United States to decide each reply, in both modes today. If your data has to stay in India end to end, a model hosted in India or on your own endpoint is available as a custom engagement.
Managed
We run it end to end. Fastest to start.
Your cloud
Same codebase, your account, your keys.
What your security reviewer will ask
Registering the calling number and having a lawful basis to call your list sit with you. We help with the first and confirm the second with you in the pilot scope.
RedEnvoy speaks Hindi, English and Hinglish, and runs on Indian providers today. If you need another language, or processing in another country, talk to us. It is a custom engagement, not a setting, and we will tell you honestly whether it is ready.
How we built it
A voice agent turns speech into text, decides what to say, turns that back into speech, and plays it down the line. Each step takes time, and the total is the pause the customer hears. RedEnvoy is built to reply in under 1.2 seconds at the median and 2.0 seconds at the slow tail, because the slow tail is what customers remember.
People interrupt, trail off and change their minds. The agent stops talking when the customer starts, and picks up where the conversation actually is. This is built into the voice engine, not added on top.
Every call is recorded from the start, and every call that misses its goal has a reason attached. That is what lets us tell you why a campaign underperformed, and fix it.
Design targets. The breakdown by stage (speech in, reasoning, speech out) is published here once it has been measured on live pilot calls.
Target · median
< 1.2 s
Target · slow tail
< 2.0 s
What we built
Latency measured at the median and at the slow tail, because the tail is what customers hang up on.
The agent stops when the customer starts talking, and picks up where the conversation actually is.
Exactly one reason attached to every call that does not reach its goal.
Calling hours, suppression lists and the recording notice are checked by the product, and a campaign that fails them does not start.
Speech-to-text, text-to-speech, the language model and telephony each sit behind an interface. Changing a provider is an adapter, not a rewrite.
The same codebase can run single tenant in your cloud account, with your keys, deployed from one configuration file.
FAQ
We test every deployment on a clean environment before your review, not during it.