pi_22by7

You do use AI! And I do too, I started my work with AI about 2 years after you did, and the experience and opinions I have match yours to a T. However, I have tried AI to do things from scratch multiple times, and the fact still remains that you have to gently nudge it like you would a junior dev. This is not a new discussion but its still worth mentioning. About music generation, I feel nervous when AI and art intersect, might be because of my background and art hobbies. Finally, boring AI is probably the best AI out there. Silently ironing out the issues and friction we face in our day-to-day without really being very flashy and glam. Also worries me that we are calling everything AI lately even if its ML or just a smart algorithm. Good for marketing, I guess.

The key insight about bloom filters lacking synergy is excellent. The ~7K document crossover point makes sense because inverted indexes amortize dictionary storage across all documents while bloom filters must encode it linearly per document

Hey HN, I'm the author.

I got tired of re-explaining my codebase to Claude and Copilot every session. They forget everything—architecture, patterns, conventions. It's like talking to a goldfish.

So I built In Memoria: an MCP server that gives AI assistants persistent memory. It learns your coding patterns from your actual code (via tree-sitter) and provides that context to any MCP-compatible AI.

Demo (2 min): https://asciinema.org/a/ZyD2bAZs1cURnqoFc3VHXemJx

How it's different from RAG or Cursor rules: - Not just retrieval: It learns patterns statistically ("90% of your functions return Result<T>") - Not static config: Learns from actual code structure and usage - Local-first: Rust core for speed, everything stays on your machine

Since launch: 86 stars, 14 forks, and yesterday we got 2 PRs for PHP support from external contributors I've never met. People are building features I haven't had time to implement, which is wild.

Works with Claude Desktop, Claude Code, Copilot, Cursor, anything that supports MCP. Install via `npx in-memoria server`.

What's the most annoying thing your AI assistant forgets? I'm prioritizing the roadmap based on real pain points.

Nice work on the game! The recording shows it off well. Good thinking using asciinema for this kind of thing.

I'd probably be the idiot who types pacman and wonders why it's trying to update my system instead of starting the game.

This looks really useful for documentation. The text-based approach seems like a huge advantage over screen recordings. Being able to copy-paste commands from a "video" is brilliant. The privacy concern about auto-uploading is a bit concerning though, glad they added the prompt to choose local vs upload.

I’m with @NitpickLawyer on this; the scan is cool, but I was hoping for a bit more detail on what’s actually being exposed. Even something like a breakdown of common tool names or clustering by function type would’ve been really insightful.

That said, it’s still surprising (and a little funny) to see how fast these things end up public. Probably lots of default setups left running without realizing they’re wide open.

oh, for sureee

Thanks for jumping in, that’s actually really helpful context. Parallelism and selective test execution built-in sounds like a huge win, especially for CI and larger teams. I’ll admit the Scala dependency still makes me hesitate a bit for pure Java projects, but clearly Mill’s design is tackling real pain points head-on. Definitely tempted to try it out on a greenfield project.

True. Tests usually dominate the time budget. But from what I've experienced, slow builds still hurt, especially when you're iterating frequently or working across multiple modules. Those 10–20 seconds of wait adds up fast when you're in the zone and trying things out.

So it did the same work that a sitemap does? Interesting.

Or maybe more like the opposite: robots.txt told bots what not to touch, while sitemaps point them to what should be indexed. I didn’t realize its original purpose was to manage duplicate content penalties though. That adds a lot of historical context to how we think about SEO controls today.

Honestly, I'm getting tired of the endless parade of "better" build tools. Maven works fine for most of what I do, even if it's occasionally painful. Gradle is... well, Gradle has its moments.

That said, Mill's performance claims sound interesting. If it really can cut build times by 3-6x, that's really amazing. I've wasted too much time in my life waiting for builds, especially on larger projects.

The Scala thing is a bit of a turn-off though. I get that you don't need to write Scala yourself, but now I'm dragging in the Scala ecosystem just to build my Java project? Feels heavy.

The IDE integration sounds nice in theory - being able to actually navigate and understand your build in IntelliJ would be pretty sweet.

Still not sure the switching costs are worth it unless you're really hitting Maven/Gradle pain points. But if I was starting a new project from scratch, might be worth a look.

This is a sharp dissection of ‘inevitabilism’ as a rhetorical strategy. I’ve noticed it too: the moment someone says ‘X is inevitable’, the burden of proof disappears and dissent becomes ‘denial’. But isn’t that framing itself... fragile? We’ve seen plenty of ‘inevitable’ futures (crypto, the Metaverse, even Web3) collapse under public pushback or internal rot.

The question I’m left with: if inevitabilism is so effective rhetorically, how do we counter it without sounding naïve or regressive?

I built this because I was tired of songwriting apps with subscriptions and ads interrupting my creative flow.

It's open source, works offline, and syncs across all platforms. You can try the web demo at flowrite.pi22by7.me

Happy to answer questions about the tech stack or get feedback!