Here's an interesting read on the evolution beyond RAG: https://www.nicolasbustamante.com/p/the-rag-obituary-killed-...

One of the key features in Claude Code is "Agentic Search" aka using (rip)grep/ls to search a codebase without any of the overhead of RAG.

Sounds like even RAG approaches use a similar approach (Query Generation).

The article raises several interesting points, but I find its claim that Claude Code relies primarily on grep for code search unconvincing. It's clear that Claude Code can parse and reason about code structure, employing techniques far beyond simple regex matching. Since this assumption underpins much of the article's argument, it makes me question the overall reliability of its conclusions a bit.

Or am I completely misunderstanding how Claude Code works?

> I did not invest the time to experiment with my SSB receiver and therefore never got any weather reports during the voyage.

wow

>But consider a personal finance app. The main goal is cross device sync, cloud backup, offline capable, etc. In this case having the entire db stored on every device is probably what you want.

A bit confused by this. If I'm a developer of a PFM, I don't want anything but a single user's financial data synced to their device. This sounds like partial replication to me.

Precisely. In the SQLSync model - every user would have a private database just for their data. For example, this is how the todo list demo works: https://sqlsync-todo.pages.dev

(Note: currently SQLSync's server tier doesn't support auth, just random 128bit ids. Auth will come as it matures - but don't use this for anything super secure at the moment).

I tried to link Splitwise, but keep getting "The authorization flow was cancelled" While the authorization window is open.

Although it seems to have actually linked successfully after going through the Splitwise auth flow despite this error.

Thanks for the report - I'll take a look at see what's going on. (There's some postMessage funny business there that tries to make the oauth experience nice that might be the culprit...)

Very cool, I signed up for a trial to kick the tires!

I created a tool that helps people stay on top of shared transactions (https://reconcile.app/) a few years ago and have recently been working on another iteration of it that has some similarities to what you've built, but with a focus on couples who share much of but not all of their financial lives.

Woah, neat! It seems like we had pretty similar thinking for the kinds of workflows you'd want.

It looks like you have an Venmo integration ("Sent to Venmo" in screenshots) which I haven't quite cracked yet - curious if you're willing to talk about how it works. Tender has venmo via plaid, but it only supports importing tx, not initiating payments/requests.

I thought about either: integrating unofficially (https://github.com/mmohades/VenmoApiDocumentation) or just doing links to venmo payment (i.e. venmo.com/user?pay=500) and having some way to import venmo contacts

Neither seems great from a UX or security perspective.

Yeah, my solution is just using a deep link - I'm unable to track whether or not the payment is completed due to the lack of a Venmo API.

I wrote about how it works here: https://gabeoleary.com/posts/venmo-deeplinking-including-fro...

When I wrote this it only worked on mobile. The deep link would take one into the app with the amount/details etc prefilled out. Just tested and it looks like the same format works on desktop now too.

I had until just now not realized that Plaid has a Venmo integration.

I could imagine using the deep link to begin the transaction and then using the plaid connection to look for it's completion...

I don't know about being able to paste in Playright steps, but a friend of mine has an extension built around extension based browser automation: https://browserflow.app/

You can record steps & have the extension replay them on your machine or in the cloud (presumably using puppeteer/playwright).

This is very cool! I was messing with some browser automation (Playwright) via GPT recently.

One idea I had: it would be cool if I could teach the agent. For instance, give it a task, but if it struggles, just complete it myself while the extension observes my interactions.

Perhaps these could be used as few shot examples for priming the model?

Gonna play around with this soon!

Check out https://arxiv.org/abs/2303.11366?s=09 ;)

We aren't introspecting previous runs or human examples to optimize workflows right now, but it's a powerful tool and one that I expect we'll employ in the future to make Taxy more efficient and reliable.

I use https://floorplanner.com for similar purposes (I pay $5/month). It is great at getting the floorplan nailed, but leaves a bit to be desired when trying to fill the space with accurate furniture, patterns etc. You can have multiple variations floor plans per level.

My gf uses https://foyr.com once we have the floor plan nailed to get an idea about what colors, textures & furniture to use in a space. It supports higher def renderings. It's kind of expensive though iirc.

Sweet home 3-D is free. Does everything that you’re paying for and will still be around in 10 years.

I used floorplanner for my last remodel. It was great. The contracting company was really impressed by what I gave them.

I'd say it will take a few hours to get used to it and after that you can probably do most of what you want to do.

Impressive generated voices for TTS

Still nothing comes close in terms of "install/usage accessibility" compared to the terribly sounding espeak