You helped us see Draft more clearly

First: thank you.

It's our first launch. Appreciate all the kind support and sharp comments. We learned a lot.

Here are the 4 themes we heard:

1. Multi-AI capture has to earn trust

People do not live inside one chatbot anymore. But multi-AI capture is only valuable if users can trust its reliability. A great question came up: what happens when ChatGPT or another AI product changes its DOM?

That is real. Our parser is AI platform-specific, designed to preserve rich content: Math, SVG, toggle lists and structured answers should survive the move from chat to knowledge base.

So reliability is now part of the product promise, not just an implementation detail. We are prioritizing automated capture tests so we can detect breakage faster and reduce user impact on the capture experience.

2. Local-first needs clearer data safety

One commenter rightly pointed out that browser-local data can be risky if people do not understand backup.

We agree. Draft supports local backup/restore. We are working hard on two options:

1. file-based workspace, directly saves to the local file system instead of keeping in the browser-local.

2. bring-your-own-storage, like Google Drive so you can keep local-first control with stronger recovery and data sync.

3. Capture should be light, retrieval should be smart

One of the best insights was about user signal. AI chats become noisy fast if everything is saved.

That changed how we think about the product. Draft should help users capture the parts worth reusing, not blindly archive every conversation.

A commenter put it well: saving should not become homework. At capture time, users should not have to tag, clean up, categorize, and explain every saved answer.

But at retrieval time, context matters. You do not just want “that answer I saved once.” You want “the answer I meant to cite,” “the claim I needed to verify,” or “the template I wanted to reuse.”

So we are exploring lightweight capture-time signals, like save-intent chips, selected highlights, or thumbs up/down, that can make capture lighter and retrieval smarter without adding organization debt.

4. Valuable knowledge gets trapped in more than chat

One commenter pointed to a related problem: useful insights also get buried in podcasts and long-form audio.

That feels like the same pattern:

The value is not just in summarizing the whole conversation, but in capturing a specific moment with enough source context to revisit it later.

Audio / Video capture is not on the immediate roadmap, but it is an exciting candidate for our future direction: turning valuable moments from long-form conversations into reusable knowledge.

-------

What's coming:

We are going to address those feedbacks in two weeks.

Meanwhile,

let's stay connected.

Tell me what you like,

what you hate,

any question,

or just say hi.

https://x.com/toliuweijing

Poll Question:

After you save a useful AI answer, what do you usually want to do with it next?