Overview
In March DuckDuckGo launched Duck.ai, a privacy‑focused chatbot platform that lets users query third‑party large language models (LLMs) such as OpenAI, Meta, Anthropic and Mistral. After adding image generation, the latest addition is a real‑time voice mode.
How Voice Mode Works
When a user starts a voice chat, Duck.ai creates an encrypted relay to an OpenAI model. The microphone audio is streamed live to OpenAI, which transcribes the speech, processes the prompt and returns both a spoken and a text response.
- Audio is sent in real time; no local recording is kept.
- The OpenAI model generates the reply and streams it back to the user.
- The connection is encrypted end‑to‑end, preventing DuckDuckGo from reading the raw audio.
Privacy & Data Handling
DuckDuckGo’s core principle—“privacy by default”—extends to voice mode. Chats are anonymized and never used to train the underlying AI. DuckDuckGo states that the streamed audio “is not stored by DuckDuckGo or by OpenAI… after the chat ends.” OpenAI is contractually limited to using the data only to perform the service.
Enabling and Disabling Voice Mode
Voice mode is optional. Users can turn it on from the Duck.ai interface and can disable it at any time via Duck.ai Settings, ensuring full control over their data.
Future Outlook
With voice, image, and text capabilities, Duck.ai aims to provide a fully private AI assistant while keeping the option to opt‑out of any feature. The company signals that more AI‑driven tools may follow the same privacy‑first approach.