item007

Super helpful！Thank you for sharing your experience; it serves as a valuable reminder for me.

Could you tell me what the main and specific challenges or difficulties were during your implementation process?

If you could fix one limitation of the current Obsidian/PKM plugin approach, what would it be (UX, reliability, on-device, or integration with tasks)?

Your needs are indeed quite similar to my own ideas.I put an optional feedback form in my HN profile, if you are open to that.

It seems you do have some pain points and needs. But is mere retrieval really enough? What I mean is, retrieval is an active pull action—you need to know what you're searching for in the first place. My goal is more about leveraging AI for push notifications to maximize your existing knowledge base. If you're interested, feel free to check out my profile to help me gather more insights.

Thanks a lot for your comments and glad to hear that my idea has inspired you. "Retrieval at the right moment" is exactly what I’m trying to solve.

Also on hallucinations: would you trust suggestions more if each proposed action came with a quoted “evidence snippet” + source link?

If you are open to it, I put some context + a short survey in my profile if you want to take a look.

This is truly a golden line—thank you so much!

This “push and pull” framework and the perspective that “noise equals insight quality” are precisely the core constraints I wanted to center my design around.

Two follow-up questions:

1.If the connector adopts a chat-first mode (similar to Signal/SMS), could this generate excessive noise? Since human input often carries emotion and subjective bias, my original intent was for the AI to serve as an emotionless, relatively neutral bridge.

2.Regarding trust mechanisms: Before implementing stricter governance measures (auditing/penalties), should we establish foundational safeguards through local-first storage + explicit export (md/json/sqlite)?

If you're open to deeper discussion and love to explore this further. I put additional information and an optional feedback form in my HN profile.

Update: I posted a short synthesis of the main themes here: https://news.ycombinator.com/item?id=46833444

Update (31st. Jan. 2026) Recurring themes/pains from replies so far (tl;dr)

1.On-demand recall & retrieval is the core pain: people capture a lot but can’t reliably resurface the right note/link at the right time; they want stronger search (fuzzy/semantic), snapshots/context, and “pull-based” recall when needed.

2.Privacy/local-first is a hard requirement for many: “no cloud, no third-party access,” ideally open-source and self-hostable; any AI must run fully on-device to be trusted.

3.Low-friction matters more than perfect organization: users prefer systems that don’t force structure or add maintenance overhead—messy-first, iterate only when a real problem appears.

4.Avoid interruption by default: many dislike proactive “AI suggestions”; they want controlled resurfacing (opt-in prompts), not constant nudges.

5.Different goals coexist: for many, notes are for memory/inspiration/reference (not turning into tasks), while others want action workflows—tools should respect both modes.

6.Cost and scalability must be predictable: long-term indexing (years of notes/history) can get expensive; pricing needs to be transparent and not “per task,” and context signals across tools are often noisy/unreliable.

I feel this. One thing that helped me is separating “notifications” into 2 buckets: (1) interrupts (things that must break my focus) vs (2) queues (things I can review on my own schedule). Most apps only offer more interrupts. What I actually want is a daily/weekly review queue that surfaces the few items that map to active priorities.

Curious: what are the top 2–3 categories you never want to miss (people, deadlines, bills, health, etc.)? And where do they originate (email, calendar, Slack, SMS)?

If you answered Q1–Q3 above, that already helps a ton. I also made a 3-min survey to aggregate patterns. I’ll share a summary of results and invite early testers if there’s real demand: https://concerns.vercel.app