Lucidos

Builds and runs the apps and automations you describe. Desktop and phone.
Open Source MIT

[Lucidos @lucidosdev]

I built an AI OS that builds its own tools.

Describe an app (e.g. track my workouts) and a minute later it's running in your workspace. Own UI, own storage, own charts. On your phone too. No deploy, no hosting, no service to spin up.

[Lucidos @lucidosdev]

I’m thinking of the agents I manage now as a consultant I had on my team a few years back. My colleagues thought he was among the best programmers in the country. He was at least twice as expensive as the other devs on the team, and fixed issues fast. But it was all short sighted patches to issues mostly caused by himself. He left behind a mess and years of extra work for us, still most didn’t understand it was his mess we were cleaning up.

People complaining about how models mess up are probably blind to the techniques needed to tie these tornadoes down. I was not, but all I managed to do in that team was being unpopular. With the agents I have my way.

[Lucidos @lucidosdev]

@VictorTaelin Linked list discussions for Elm from back in the day came to mind, then I found somewhere you've thought about supporting Elm for Bend? That would be sweeet!

[Lucidos @lucidosdev]

@claudeai Or just use Lucidos - create *any* app, and own all of your context on your own devices

[Lucidos @lucidosdev]

Thanks again, man! I'll let you know when our release is out. There were other complications creating the "Tree memory" module and the switching from "Classic memory" in Lucidos, so my initial estimate for this was a bit off.

My product manager at my real job would not have been surprised by this development

[Lucidos @lucidosdev]

@mattpocockuk keep coming back to this for all augmentation work; using others' approach as inspiration and adapt to team/project needs

[Lucidos @lucidosdev]

@VictorTaelin that'd be awesome, thanks, I will!

[Lucidos @lucidosdev]

When I see a bug or something I don't like, I always ask "What can I do to enforce this never happening again?"

Then fix the category error structurally, mainly using one of these:

1. Refine domain model to make invalid states impossible
2. Enforce test regimes with rules applying to different types of files
3. Use hooks to run scripts enforcing e.g rules about comments
3. Code generation (e.g Rust -> Typescript) to make drift impossible

Or go even further and use something like Acadia for DB and a proof language like Bend

[Lucidos @lucidosdev]

@karpathy You need a home for all of this - come to us

[Lucidos @lucidosdev]

@unclebobmartin And to keep the code in good condition to continuously evolve the product? That's my main reason, and why I have agents doing it still.

[Lucidos @lucidosdev]

@poteto That’s beautiful and relatable. Made Lucidos for anyone to experience the joy of building!

[Lucidos @lucidosdev]

@mattpocockuk If you want more on 1) I recommend Elm (eg the talk Making impossible states impossible) and Acadia to go all the way to db with awesome type system.

And Bend, already in her own reply

[Lucidos @lucidosdev]

@poteto @bot want to take the principle of categorically impossible states to the db as well? check out this: https://t.co/EgUwW5ygBu

[Evan Czaplicki @evancz]

Run an Elm program as a simple HTTP server!

✨ https://t.co/6lg8CMIU5P ✨

Check out the full client/server/database example written in Elm/Acadia, and please share your results if you evolve it into something cool!

Very curious to see what you can make!

[Lucidos @lucidosdev]

@tobi We call it "If you can describe it, it exists" - run your automations and apps while talking to them

[Lucidos @lucidosdev]

@DanielSmidstrup We got you, come check us out

[Lucidos @lucidosdev]

@dhh You can orchestrate codex and other agents from other software. Makes sense to have a second layer for this anyway

[Lucidos @lucidosdev]

@thekitze Never understood how anyone would want to talk to their agent through one of those chats. Why put on a straitjacket when you want to run?

Need seamless back and forth between chat and custom apps. On the phone, no telegram.

[Lucidos @lucidosdev]

@thekitze Yeah, some of us already live in a unified experience that knows the user. Research, building and running apps and automations all part of the same surface

[Lucidos @lucidosdev]

And the more they know about *other* domains, the more useful they are as well. If you hired a frontend engineer, would you restrict it from knowing good UX and backend as well? I'd rather promote them to product engineer and get work done faster.

The one reasonable argument I've heard is context management, but I'd rather handle that dynamically than restricting it to a static agent config.

[Lucidos @lucidosdev]

A good mix of orchestration threads and sub-threads does it for me. Depends on the task, smth like a series of security fixes can be lead by one parent, where sub-threads report back, the parent escalate to human if necessary. So some of these and maybe one or two deep work new feature tasks, plus some bug fixes works for me.

All organized in a thread pane with sub-threads. Can switch to needs attention/review view