Catch up on unread DMs with AI

Grais is a desktop browser extension, so it works through the threads exactly where they are: WhatsApp Web, Telegram Web, Instagram, LinkedIn, and Tinder. Open a conversation and it reads the history it can see, summarises what the last messages are actually asking for, and drafts an answer. Nothing has to be forwarded, pasted, or re-explained, which is the step that makes a forty-thread backlog take an afternoon in a general-purpose chatbot.

The part that compounds is memory. Grais keeps context per contact across conversations, channels, and months, so a thread you last touched in June does not start from zero. It also tracks what was promised in both directions. When you are behind, that list is the real backlog: not the unread count, but the three people waiting on something you said you would send.

• It says what the thread needs. Each conversation gets a reading of intent and stakes from the visible context, so triage does not mean reconstructing the thread from the top.
• It flags what went quiet. Grais surfaces threads where the other person stopped replying and proposes a way back in built from what was said before.
• Approving is the whole loop. Drafts wait for you by default. Trust is set per contact, so the twenty threads you would never automate stay manual.

One message is not a problem worth solving. Anyone can answer one message. The job here is the fortieth, where each reply carries a fixed setup cost — who is this, what did we agree, what did I already say — and forty setup costs is the whole evening even though the writing itself would take a minute per thread.

Pasting a thread into a chatbot removes the writing cost and leaves the setup cost in place, because you are the one supplying the context every time. Grais removes the setup cost instead, which is the larger half. That is the entire difference in this use case, and it is worth being blunt that it is the only difference that matters: the drafts are not better than what you would write, they are simply already there.

The practical shape of a catch-up session is: open the platform, walk the unread list, read the one-line reading of each thread, approve or edit the draft, move on. Threads that need real thought get pulled out of the run and given the time they deserve, which is easier to do once the other thirty-five are gone.

Unread counts are a bad measure of what you are behind on. Most unread messages need a two-word answer or no answer at all. The threads that matter are the ones where someone is waiting on something specific, and those are usually already read, already answered once, and now silently overdue.

Grais keeps that as an explicit list rather than something you reconstruct: what you promised, what they promised, and which conversations have gone quiet since. It is the difference between clearing an inbox and actually catching up, and it is the reason the memory matters more here than the drafting does.

For a thread that needs several messages to unwind — a delayed project, a favour you have to renegotiate — set a conversation goal with `/goal`. Grais maps the sequence, decision points, and fallbacks, so the first message is written knowing where the conversation has to end up.