Yeah nah thanks I'm building my own. An agent that knows every bit of my life? Dream come true for data harvesting. It's bad enough what it can infer and sell from the public statements I make in social media and imperceptible metrics I could not fathom.

Same. Personal ORM essentially with a sqlite DB that sinks to a home server. Totally useful without an agent at all, but exposes basically all functionality via tools that have access limited for private information and the like.

My favorite part is that I'm in full control. All that caution about giving an app phone permissions? Gone. I could track my physical location in a table recording all the DB changes if I wanted to.

All it's missing is a local model.

This is where we're at, and all us devs are taking a crack at building our own. Then some of them take off like OpenClaw or Hermes but I don't find myself particularly drawn to those solutions. Its still very hackerish.

Also the swiss army knife approach feels wrong. Too much choice. It can connect to everything, ok great, so I could chat with it on telegram, discord, email, whatsapp, etc. For a person who comes along and wants to configure great. But for someone who just wants to use it and get a web ui and maybe some sane defaults it's a bit overwhelming. If we can go from super highly configurable does anything to sane defaults with an opinionated interface and its open source we might get somewhere. Not only that but if we can get to like Proton/Notion sort of equivalent alternatives where everything is not controlled by one player that would be good. Proton did mail. Notion did docs. Maybe someone else will do chat.

I agree I want a self hosted thing.

I can't wait until we have drastically better models we can run locally that don't cost as much as a car.

Garbage being created at unprecedented rates does not diminish the fact that LLMs are acting as a positive multiplier for useful tools; this very item being one.

No one in all these years had gone through the effort of replicating the Spotify client with a snappy fast interface for a reason. Everything had been an integration into already existing software, or simpler interfaces, because the effort was not worth it. The equation has changed.

Wayland is stable. What it's not is stagnant, like X11, which lacks fractional/mixed-DPI scaling, per-window HDR/color management, explicit GPU synchronization, and a security model ready for the modern challenges of the mid 90s.

It's also technically cross-platform, as evidenced by FreeBSD's support for Wayland compositors. But it's not its goal either, the goal is an excellent Linux desktop, not something that serves the 17 people in the world that routinely run the same exact desktop config on Linux and non-Linux PCs.

> What it's not is stagnant, like X11

X11 used to crash all the time, especially after an update, back when it was being vibrantly developed. It's been a lot more stable since it got abandoned. I've not had a crash in 13 years.

The DPI thing is nuts, any high res monitor these days is unusable without some DPI adjustment.

With Pi as a harness I've been using OpenAI models as a worker with an Opus 5 (in Claude Code) planner. I've only had a few issues with Terra High/Medium and absolutely none with Sol Medium+ on a fairly complex Rust project that targets Linux, Mac, Windows and Web, with plenty of nasty FFI, VMs, remotely debugging systems, among some other things within a monorepo.

I think the key is to give them a nice assortment of self-verification tools, an AGENTS.md or reference document that they're encouraged to routinely check, and asking the planner to be thorough with the ACs but give the model some space.

The planner routinely finds issues with the worker's output, but that's what it is for.

Harness is the key. I built my own to "talk" our institutional knowledge and it's working great

Plan with sol-med, implement with luna-high. Rarely a problem.

Same for me, with Claude Opus/Sonnet. All the models are almost equivalent if well steered

This ignores that whenever the margins for an industry increase, this interests new investments in it, and new players to come aboard, as well as new product lines that can satisfy the demand in a new way that people either didn't consider or rejected before due to being outclassed by the formerly cheaper alternative.

Rationing does not stimulate investment this way.

I never said rationing does anything else except spread around inventory to many customers.

The overwhelming majority of the enjoyable coffee experiences are caffeinated. While there is good decaf out there it's not the norm, specially in smaller markets.

I think they meant that coffee contains a lot of other compounds than just caffeine, which something like energy drinks or teas will not include. So you can't necessarily extend conclusions from a study on consumption of coffee to effects that other drinks that happen to include caffeine might have.

Edit: this is especially relevant here, as the study found similar effects in decaffeinated coffee drinkers. So the effects they observed, if real, are not related to caffeine.

I've had success asking it to specifically spawn a subagent to evaluate each work iteration according to some criteria, then to keep iterating until the subagent is satisfied.

I’ve had great success replacing it with Kimi 2.6

Is there Live Edit support? I've literally left an agent running to automate the use of Android Studio out of my workflow when I just want to iterate on UI.

Look up protondb for game compatibility. Arc raiders is marked as running perfectly fine, but plenty of multiplayer games with invasive client side anti cheats such as Fortnite or Genjin Impact do not run. If you depend on such games it's best not to switch, any privacy concern you may have with Windows is gone with those games running literal rootkits on your PC anyway.

I've been running Linux for gaming for well over 15 years and have not missed much in the last 5 or so. There's way too many games out there to play that do run on Linux even if unemployed and have the time to dedicate it as your sole hobby.

If you generate OpenAPI specs, and clients, and server type definitions from a declarative API definition made with Effect's own @effect/platform, it solves even more things in a nicer, more robust fashion.