What sort of context do you give the APIs when you are starting the game? Does it need to learn the rules as it goes?

We have a standard harness for each of the model's that we test. Each prompt includes the rules, access to memory, and a lookup of the complete ruleset. The prompt adapts adding legal actions per turn and guidance depending on the stage of the game (updated based on the technological progress of the player).

Unlike RL algorithms these LLMs wouldn't learn quick enough without the prior knowledge the harness provides

what do you use for memory?

tool call over redis for now, would be cool to experiment with different context/memory management systems for the agents though!

an unbelievably cool style — excited for new generations to "discover" him

thanks for catching this. just fixed.

note that I also have a system where if the temperature seems outlier compared to direct neighbors it averages the 3 nearest neighbors. this usually occurs in neighborhoods with a single sensor that can skew the results heavily at certain times of the day, etc.

Good flag. I've just added add fallback to the nearest location with a sensor to the repo.

This was directly inspired by Mr Chilly which was designed by my friend Anna Bleker.

It's an excellent iOS app: https://mr-chilly.com/

My goal was to do something similar as a Claude Code skill

I use Mr Chilly to demonstrate to non-SF folks how many microclimates SF (and the Bay Area has).

Only suggestion: separate Inner and Outer Sunset since there can be a massive difference between near Ocean Beach and near Irving/9th Ave in autumn (ie. SF's hottest season).

Edit: nevermind, just saw both inner_sunset and outer_sunset in /neighborhoods. I'd assumed it was merged based on the human readable list on the landing page. Thanks for the fun API!

thanks I will update the homepage to reflect this.

submit a PR

no I made this primarily for a Claude Code / Clawdbot skill so I am not making it super sophisticated.

You should use Purple Air if you want to make it more focused https://www2.purpleair.com/

> I don't understand why it includes indoor sensors at all let alone by default.

Add location_type=0 to only get outdoor sensors

or just click the buttons that accomplish the same thing. The point is someone at PurpleAir is asleep at the wheel if such an obvious default configuration isn't being set. If they can't get such a basic thing right, why do we trust anything else from them? "Anything else" specifically including "running their software on a raspberry pi inside my home network".

Because PurpleAir is not a weather service, they’re a sensor company.

Fork the Github! Would love to see it elsewhere :)

I made this primarily to use inside Claude Code in terminal but maybe I'll make a little demo on the website if you put in an SF zip code.

Demo added: https://microclimates.solofounders.com/all