Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A hackable AI assistant using a single SQLite table and a handful of cron jobs

geoffreylitt.com

Post reply on HN
Print view View original

161–170 of 183 posts

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#161

Post by n_ary »

sunshine-o wrote: post #34

This is brilliant ! I am wondering, how powerful the AI model need to be to power this app? Would a selfhosted Llama-3.2-1B, Qwen2.5-0.5B or Qwen2.5-1.5B on a phone be enough?

Having some experience with weaker models, you need at least 1.5B-3B to see proper prompt adherence and less hallucinations and better memory.

Also models have subtle differences, for example, I found Qwen2.5:0.5B to be more obedient(prompt respecting) and smart, compared to LLama3.2:1B. Gemma3:1B seems to be more efficient but despite heavy prompting, tends to be verbose and fails at formatted response by injecting some odd emoji or remark before/after the desired output.

In summary, Qwen2.5:1.5B and LLama3.2:3B were the weakest model which were more useful and also includes tools support(Gemma does not understand tools yet).

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#162

Post by simonw »

fmbb wrote: post #155
xp84 wrote: post #122

Earlier quoted context omitted.

The Apple walled garden argues against you here. There are at least 20 million families in America where this holds true: • Everyone in household uses an iPhone • Main adult family members use iCloud Mail or at least use Apple Mail to read other mail • Family members use iCloud contacts and calendars • USPS Informed Delivery could be used (available to most/all US addresses) • It can be ascertained what ZIP code you'…

The thing that is standing in their way is probably that nobody is willing to pay for this what it costs to run.

Doesn't look very expensive to me. An LLM capable of this level of summarization can run in ~12GB of GPU-connected RAM, and only needs that while it's running a prompt.

The cheapest small LLMs (GPT-4.1 Nano, Google Gemini 1.5 Flash 8B) cost less than 1/100th of a cent per prompt because they are cheap to run.

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#163

Post by hm-nah »

xp84 wrote: post #119

I don't know if I love this more for the sheer usefulness, or for the delightful over-the-top "Proper English Butler" diction. But what really has my attention is: Why is this something I'm reading about on this smart engineer's blog rather than an Apple or Google product release? The fact that even this small set of features is beyond the abilities of either of those two companies to ship -- even with caveats like "…

It’s because this story hints at the concept of “Unmetered AI”. It can be easily hosted locally and run with a self-hosted LLM.

Wonder if Edison mentioned Nikola Tesla much in his writings?

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#164

Post by orangebread »

sci_prog wrote: post #132
IanCal wrote: post #81

Earlier quoted context omitted.

Also an email that comes back a minute later feels fast. A chat that types at the same speed feels slow.

I'm building something similar. See my comment the OP above: https://threadwise.app

If you want to get ahead of the curve, look into the Agent-to-Agent protocol Google just introduced. I'm currently using my own custom AI agent assistant to perform life tasks. If I could integrate a better tooling/agents into my own assistant system like your's that'd be awesome.

It's kind of like sure, I could manage my own emails, or I could offset this to someone who does it better. If you do it better and it's affordable, I'm in.

We are on that starship to the future right now and I love it.

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#165

Post by orangebread »

gklitt wrote: post #54
maxmcd wrote: post #35

Earlier quoted context omitted.

This project has a pattern just like that to handle the inbound USPS information: https://www.val.town/x/geoffreylitt/stevensDemo/code/importe... I think it would be pretty easy to extend to support other types of inbound email. Also I work for Val Town, happy to answer any questions.

yeah i actually do handle inbound email! just forgot to include that code in the shared version. the telegram inbound handler shows the rough pattern.

is there a reason you went with telegram and not slack or discord? i was thinking that it could open up a broader channel for communicating with your assistant. i understand you're also just building more of a poc, but curious if you'd thought about that. great work btw :)

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#166

Post by charlieyu1 »

angusturner wrote: post #154

The thing this really hits home for me is how Apple is totally asleep at the wheel. Today I asked Siri “call the last person that texted me”, to try and respond to someone while driving. Am I surprised it couldn’t do it? Not really at this point, but it is disappointing that there’s such a wide gulf between Siri and even the least capable LLMs.

Siri poped up and suggested me to set a 7 minute timer yesterday evening. I think I did it a few times in the week for cooking or something. This is a pretty stupid suggestion, if I need it I would do it myself.

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#167

Post by kylecazar »

philsnow wrote: post #126
kylecazar wrote: post #50

I like the idea of parsing USPS Informed Delivery emails (a lot of people I encounter still don't know that this service exists). Maybe I'll make something to alert me when my checks are finally arriving!

This part was galling to me; somewhere in the USPS, the data about what mailpieces/packages are arriving soon exist in a very concise form, and they templatize an email and send it to me, after which I can parse the email with simple+brittle regexes or forward the emails to a relatively (environmentally-)expensive LLM or so.... but if they'd made the information available with an API or RSS feed, or attached the json…

It would indeed be nice to have a recipient/consumer-side API!

I don't think it'll ever happen. Really the only valid use-case would be for people to hack together something for themselves (like we are discussing)... They don't want to allow developers to create applications on top of this as a 3rd party, as informed delivery itself has to carefully navigate privacy laws and it could be disastrous.

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#168

Post by mikethemerry »

Along the same lines, I've just done a build called Jeeves. A bit less flair, but very fast to put together. The stack is:

1. Claude Desktop 2. Projects 3. MCPs for [Notion, Todoist] and exploring emails + WhatsApp for a next upgrade

This is for me to support productivity workflows for consulting + a startup. There are a few Notion databases - clients, projects, meetings, plus a Jeeves database. The Jeeves database is up to Jeeves how it uses it, but with some guidance. Jeeves uses his own database for things like tracking a migration of all of my previous meeting notes etc under the new structure.

So my databases, I've set up my best practices for use. Here's how my minutes look, here's how client one pagers looks like, here's the information to connect it all together, and here's how I manage To Dos. I then drop in transcriptions into a new chat, with some text-expanding prompts in Alfred for a few common meetings or similar, and away he goes. He'll turn the transcript into meeting notes, create the todos, check everything with me, do a pass, and then go and file everything away into Notion and Todoist via MCP.

It's also self documenting on this. The todoist MCP had some bugs, so I instructed Jeeves to go, run all the various use cases it could, figure out the limitations and strengths, document it, and it's filed away in the Jeeves database that it can pull into context.

It lacks the cron features which I would like, but honestly, a once-a-day prepared prompt dropping into Claude is hardly difficult.

Re: A hackable AI assistant using a single SQLite table and a handful of cron jobs

#169

Post by bronco21016 »

xp84 wrote: post #119

I don't know if I love this more for the sheer usefulness, or for the delightful over-the-top "Proper English Butler" diction. But what really has my attention is: Why is this something I'm reading about on this smart engineer's blog rather than an Apple or Google product release? The fact that even this small set of features is beyond the abilities of either of those two companies to ship -- even with caveats like "…

As some of the other commenters have directly and indirectly pointed out, I believe this is the crux of the AI Agent problem. Each user has a customized workflow they’re trying to achieve. This doesn’t lend well to a “product” or “SaaS”. It leads to thousands of bespoke implementations.

I’m not sure how you get over this hurdle. My email agent is inevitably different than everyone else’s email agent.

Post reply on HN