I don't even need today's frontier, give me a local model I can run on my Mac comparable to Claude 4.5 as of December last year and I'll probably lose any interest in new hosted LLM advancements altogether.
2. Bun.$ (Bun Shell) to execute the macOS command to take a screenshot (I do this for all connected screens at that moment)
3. Bun.Image to downscale everything to 1x in case some of the screenshots are 2x
4. Bun Shell again to run a JXA AppleScript thing to use the Vision Framework or whatever it is called to OCR the image into a file
5. Bun Shell to run the Swift compiler in the one-off eval mode with inline Swift helper that runs the Foundation Models Framework built-in LLM with a system prompt that tells it what the OCR said and instructs it to glean what may be on the screen (can't do this with JXA because the models are not exposed with ObjC APIs)
6. For each screenshot, continuously, take the previous day summary file and the last OCR/context results and produce a new summary of the day
I plan on adding extra information from the OS like the currently opened windows, currently focused window, time of day etc. into the mix, but so far it hasn't been needed. It produces reports of a good enough quality for me.
I `grep` these daily summaries whenever I need to recall a link I saw or a find what channel a message I spotted was in or take another look at that one tab I already closed, maybe re-open it by its OCR'd URL etc.
I have a personal, fully offline and local version of Windows Recall basically, but good, made using macOS built-in OCR and LLM. The reasoning requirements are tiny (just interpret the screen based on the OCR, do rolling de-duplication and summarization), but they are non-zero. The tool is valuable to me and it being dep-free and fully offline and local just gives me a good feeling.
If you're on macOS you can try the built in LLM which I think is similar in size. There's a project called Apfel that wraps it in a CLI. Also Chrome ships with a web API called Prompt API that gives you offline access to Gemini Nano which can do both text and images at the input. Also tiny. I've integrated these into my workflows where a tiny but non zero amount of reasoning is needed in between the otherwise fully deterministic steps.
I think Last.fm might have been a better friend finding and dating app than any of its contemporaries or anything that came after. Seems like everyone in this thread or anyone I know IRL has a story of making a good connection with someone via it. I know I'll always cherish the people I got to know on there.
There is basically no support to speak of. Scaling a zero is not hard. You can be paying 200 USD a month with barely any chance of ever hearing back. Your best chance of getting support from Anthropic is the same as with any other big tech company: have a Twitter following or know someone who works there.
I stopped checking failure emails from GitHub Actions unless they happens half a dozen times or so a day because I get more false negatives (GitHub Actions issue not my workflow issue) than legitimate notifications I can act on. I'm from Europe too.
I asked it about designing a 12 V solar system for a garden shed and it got everything but the broadest of strokes wrong. It figured out there should be a solar panel, a solar charge controller, a battery and some loads, but the wiring was non-sensical and when I drilled in on the solar charge controller settings etc. it completely fell apart. Absolute non-starter for any information you plan on depending on, but good entertainment value and impressive execution.
Same story here, I installed it and ran `gwc auth setup` only to find I needed to install a `gloud` CLI by hand. That led me to this link with install instructions: https://cloud.google.com/sdk/docs/install. Unmistakeable Google DX strikes again.
In theory, comments on Hacker News should advance discussion and meet a certain quality bar lest they be downvoted to make room for the ones that meet the criteria. I am not sure if this ever was true in practice, it certainly seems to have waned in the years I have been a reader of this forum (see one of the many pelican on a bike comments on any AI model release thread), but I'd expect some people still try to vote with this in mind.
Being sarcastic doesn't lower the bar for a comment to meet to not get downvoted, so I wouldn't go thinking people miss the sarcasm without first considering whether the comment adds to the discussion when wondering why a comment is downvoted.
IME CLAUDE.md rarely gets fully honored. I've left HN comments before about how I had to convert some CLAUDE.md instructions to pre-commit deterministic checks due to how often they were ignored. My guesstimate is that it is about 70 % reliable. That's with Opus 4.5. I've since switched to GPT-5.2 and now GPT-5.3 Codex and use Codex CLI, Pi and OpenCode, not CC, so maybe things have changed with a new system prompt or with the introduction of Opus 4.6.
TIL that HA notifications can have associated actions. I have the exact same setup as you, except I only receive the notification and then walk over to the laptop to unblock the agent feeling like a human tool call. This will improve my workflow, thank you.
Not the person you replied to, but I'll stress the point that it is not just what you can add that Claude Code doesn't offer, but also what you don't need to add that Claude Code does offer that you don't want.
I dislike many things about Claude Code, but I'll pick subagents as one example. Don't want to use them? Tough luck. (AFAIK, it's been a while since I used CC, maybe it is configurable now or was always and I never discovered that.)
With Pi, I just didn't install an extension for that, which I suspect exists, but I have a choice of never finding out.