An assistant that knows your loved ones, habits, and projects to pick back up

Based on its user’s messages, habits, and projects, Quest builds a personal memory to suggest actions without delay.

One hundred fifty-six files on his projects, interests, and personality, plus a map of 358 relationships: that is the memory Vincent Nallatamby says he obtained by linking his accounts to Quest. His personal assistant must use this context to help him in daily life and suggest actions without waiting for him to make a new request.

The founder defends “radical transparency” to obtain this kind of assistance. The privacy policy, however, states that each connection is chosen by the user. Access to health and location is disabled by default. The service draws on the information entrusted to it to track what it calls “quests”: learning something, carrying out a personal project, or maintaining relationships. A memory that also includes hypotheses

The Quest presentation highlights more than a hundred wiki-style files and a mapping of relationships. In his case, the founder says he has connected his emails, WhatsApp, and Instagram, with memory updated daily. The publisher also says that initiative-taking improves with user feedback and the agent’s mistakes.

The memory can contain facts, anecdotes, and hypotheses about personality or goals. It can be viewed and corrected in the app. Disconnecting a service does not erase the knowledge already drawn from its data: facts are not categorized by source. Their removal must be requested separately.

The benefit sought is to be able to resume a conversation or start a search without having to recount one’s background, preferences, and the people involved. The relationship map extends this context to close contacts and interlocutors present in the exchanges. It therefore concerns not only what the user explicitly says about themselves. A phone for acting in conversations

The agent has its own Android phone to use apps such as WhatsApp, Instagram, or TikTok. The app intended for the user remains separate from this device: the phone is used to carry out certain actions on their behalf.

The agent’s phone and browser run in the cloud. For messaging apps, the actions concern the selected conversations. Delegation can authorize replies without validation each time, within the limits of the instructions; the service says it asks for approval before sharing sensitive information.

In one of the examples told by Vincent Nallatamby, the assistant follows a fathers’ WhatsApp group in his neighborhood dedicated to giving away items. It can offer items on behalf of its user and alert him when an item seems suitable for his child. This account illustrates conversational delegation, with messages sent under the account holder’s identity.

According to the publisher, the data are processed in the United States, with no advertising or use for training the models. The elements needed for a task may be passed on to AI providers. Tools built around each goal

The service also announces customizable iOS mini-apps, created around the user’s quests. A goal thus becomes a tailored tracking space, in addition to the conversation.

Delegation can also go through other assistants. The founder explains that he had Quest direct Instinct to put together a selection of history podcasts. He reports nine exchanges between the two assistants before obtaining the result. Managing Muse is also among the announced features.

An App Store listing in the U.S. lists the app in English. Exchanges with Q., the name given to the assistant, also take place via iMessage, by text or voice message. Everyday events trigger the follow-up

The most concrete examples involve reusing information already available. After connecting Granola, Vincent Nallatamby says he receives an analysis of his English mistakes based on the transcripts of his calls, then a quiz to work on them. The same meetings are used to build up a bank of post ideas, drawn from what he actually said.

For travel, he says he shares his saved places in Google Maps and his live location. The assistant can then suggest visits or activities during the day, based on the tastes it attributes to him. The trigger becomes a place or an event, rather than a question asked in a chat window.