Google is shutting down Google Assistant on Android in September, and the replacement the company wants you to use is Gemini. That is fine for a lot of people. It is not fine for anyone who has started paying attention to where their voice goes after they speak it. Every command sent to Gemini leaves the phone, hits a Google data center, and feeds the same machine learning pipeline that powers the company’s ad business. If that bothers you, or you just want a voice assistant that does the handful of things assistants are actually good at without phoning home, there is an open-source app called Dicio that has been quietly doing the job for years.
Dicio is a free voice assistant for Android that processes everything locally. There is no account. There is no cloud round trip. There is no telemetry. You speak, the app converts your voice to text using a bundled speech engine called Vosk, matches your sentence against a list of known commands, runs the matching skill on the device, and answers you out loud. The whole loop happens in about two hundred milliseconds on a mid-range Android phone, and nothing about you leaves the hardware.
Why local actually matters for a voice assistant
Most people do not think twice about their voice assistant because the assistant is invisible. You say “set a timer for fifteen minutes,” the timer gets set, and you never see what happened in between. What happened is that an audio recording of your voice got uploaded to a server, transcribed there, run through a natural language model, and the response got streamed back. The audio and the transcription are now in someone else’s database, tied to your account, and used to improve the model that the same company sells to advertisers.
Dicio does the same job without that step. The Vosk speech model runs on your phone. The sentence matcher runs on your phone. The skill that handles timers, weather, calls, and translation runs on your phone. The audio never gets uploaded because there is no upload code path. The app does not have the network permission enabled for anything beyond the optional translation skill, which uses an offline model anyway.
For most people, this is the difference between a voice assistant they trust and one they tolerate. The interesting thing is how little capability you give up.
What Dicio can actually do today
There are thirteen skills in the current build. They cover the categories that account for something like ninety percent of what people use a voice assistant for in a normal day. Timers and alarms, weather lookups, calling and messaging contacts, media playback control, opening installed apps, on-device translation, and a calculator. Each one is its own little plugin with a clear set of phrases it understands, and the app tells you up front what to say.
A few are worth calling out because they are the ones that turn the app from a curiosity into a daily driver:
- Timers and alarms work the same way they did on Google Assistant, including the ability to set multiple timers at once and label them so the phone knows which one is which.
- Calling a contact by name uses the on-device contacts list. The lookup never leaves the phone, and the contact never gets sent to a server.
- The translation skill runs an offline model that handles a few dozen language pairs. It is good enough for travel phrases and basic conversation, not for documents.
- Media playback control integrates with whatever music or podcast app you already use. There is no separate media service; it just sends the standard media button intents that Android apps already listen for.
The skills are listed in a clean grid on the home screen. Tap any one to see the exact phrases it understands. There is no ambiguity, no guessing, no “I think you wanted to play music” when you meant to set a timer.
Setup that takes three minutes, not thirty
The first time you open Dicio, the home screen shows three setup cards. Tap each one in order and you are done. Microphone permission, which Android scopes to “while using the app” so Dicio cannot listen in the background. The Vosk speech model download, which is a few hundred megabytes and only happens once. The optional permissions for the skills that need them, like contacts for calling and the calendar for reminder skills if you enable them.
That is the entire setup. There is no account creation. There is no email address. There is nothing to sync, nothing to back up, nothing that depends on a service staying online. When you uninstall the app, every byte of data goes with it, because every byte lives on the phone.
The app is available from three places. The Google Play Store build is the most convenient. The F-Droid build is the one to choose if privacy is the reason you are switching, because F-Droid compiles and signs the package itself rather than trusting Google’s automated pipeline. The GitHub release is for people who want to read every line of code first, or who prefer sideloading from an APK they can verify themselves.
Trade-offs
Dicio is the right answer for a specific shape of problem. You use a voice assistant for timers, alarms, weather, calls, and the occasional translation. You do not want to send your voice or your contacts to a cloud service. You do not mind that the assistant cannot answer random questions, summarize documents, or hold a conversation.
If that is you, Dicio is the best voice assistant you can install on Android today, and it has been for a while. It is fast, the skills cover the daily cases, and the privacy story is not marketing. The audio never leaves the device. The contact list never leaves the device. The translation model runs locally. There is no analytics SDK buried in the manifest because the project does not have a reason to include one.
There are a few situations where Dicio is still the wrong tool. If you depend on smart home routines to control lights, locks, and thermostats, Dicio cannot do that, and the open-source skill list does not include a path to it. If you want an assistant that can read your email out loud or summarize a long article you just opened, you want a large language model in the loop, and that means Gemini or another cloud assistant. If you want to ask free-form questions and get free-form answers, sentence matching is the wrong architecture. The whole point of Dicio is that it does a small set of things well without the data footprint of a chatbot.
The last voice assistant that respects the user
The honest summary is that Dicio solves a smaller problem than Gemini, and it solves it more honestly. For most of what people actually use a voice assistant for, the smaller problem is the one that matters. A timer does not need a language model. A weather lookup does not need to train on your voice. A phone call does not need to upload audio to be initiated. The cloud round trip was always a choice, not a requirement, and Dicio proves that you can ship a useful voice assistant without making it.
If you have been meaning to leave Google Assistant before the September deadline, install Dicio from F-Droid this week. Spend one day using it for the things you actually use a voice assistant for. If the trade-off fits, you will not look back. If it does not, the install took three minutes and cost nothing.