Productivitat Intel·ligència artificial Programació Codex OpenAI Perifèrics

Codex Micro a prova: què aporta el teclat de 230 dòlars per controlar Codex

El creador de fal prova Codex Micro, el petit teclat d’OpenAI i Work Louder que mostra l’estat de sis agents, activa ordres i permet dictar prompts. El resultat és atractiu i útil en alguns fluxos, però clarament de nínxol.

Un comandament físic per al treball amb agents

Codex Micro és una edició limitada creada per OpenAI i Work Louder a partir del Creator Micro. No pretén substituir el teclat principal: és un comandament auxiliar que posa a l’abast tecles dedicades, un dial, un petit joystick i indicadors RGB connectats amb els xats de Codex dins l’aplicació d’escriptori de ChatGPT.

El vídeo de fal és un primer contacte, no una anàlisi de llarg termini. El seu autor deixa clar que és aficionat als perifèrics especialitzats, que va comprar el dispositiu sense haver-lo provat i que la seva eina habitual de programació és Claude Code. Aquesta perspectiva és útil perquè evita presentar Codex Micro com una necessitat: l’interès inicial és sobretot tàctil, visual i experimental.

A la botiga oficial apareix amb un preu de 230 dòlars i, en el moment de publicar aquest resum, consta com a esgotat. El paquet inclou el comandament, un cable USB-C i un joc de 32 capçals de tecla amb icones de Codex.

Què hi ha a la caixa

La primera part del vídeo se centra en el desembalatge. L’autor destaca l’acabat de la caixa, el motlle interior i la presentació, que compara amb l’experiència d’obrir un producte d’Apple. El dispositiu és compacte, amb una combinació de tecles de plàstic i sis tecles translúcides que deixen veure els interruptors i la il·luminació.

Les especificacions oficials concreten el maquinari: tretze interruptors mecànics, un sensor tàctil, un codificador rotatori i un joystick planar. La carcassa combina policarbonat i alumini, mentre que els capçals són de PBT i policarbonat. Es pot connectar per Bluetooth o USB-C i és compatible amb macOS i Windows.

A més del joc instal·lat, la capsa incorpora icones per adaptar visualment les tecles a funcions diferents: terminal, document nou, paperera i altres accions. El vídeo també comprova que la unitat arriba carregada i pot encendre’s sense cable. Per configurar altres usos i capes fora de ChatGPT es fa servir Work Louder Input.

Sis tecles que mostren l’estat dels xats

La funció més distintiva són les sis Agent Keys translúcides. Cada tecla pot seguir un xat i encendre’s amb el seu estat. La guia oficial associa el blanc amb un xat inactiu, el blau amb una tasca en procés, el verd amb una resposta acabada que encara no s’ha llegit, l’ambre amb una aprovació o resposta pendent i el vermell amb un error.

En prémer una tecla, Codex canvia directament al xat assignat. Per defecte segueix els sis xats actualitzats més recentment, però es pot configurar perquè prioritzi els fixats, els que necessiten atenció o una selecció manual. Aquesta és la part amb més sentit per a qui manté diverses tasques o agents en paral·lel: permet veure d’una ullada quin treball continua actiu i quin ja ha acabat.

La demostració ho fa visible. Mentre Codex modifica una web, la tecla corresponent queda blava. Quan la tasca acaba, passa a verd; en tocar-la, l’autor torna al xat i revisa el resultat. La millora no és la capacitat de l’agent, que és la mateixa, sinó reduir el cost mental d’anar comprovant finestres.

Tecles d’ordre, joystick i dial

Codex Micro inclou sis ordres per defecte: activar o desactivar el mode ràpid, aprovar, rebutjar, continuar un xat en una conversa nova, iniciar el mode de prémer per parlar i enviar el missatge del compositor. Les assignacions es poden canviar des de Settings > Codex Micro.

El creador del vídeo substitueix una de les funcions perquè no vol obrir tantes bifurcacions de projectes. Aquesta personalització és important: un perifèric amb poques tecles només és pràctic si cada usuari hi posa accions que repeteix sovint. La documentació actual permet vincular-hi ordres com obrir el terminal o el navegador, revisar canvis, crear un commit, preparar una pull request, adjuntar fitxers, gestionar tasques programades o obrir Skills.

El joystick transforma quatre direccions en accions. La configuració inicial activa el mode Pla cap amunt, mostra o amaga la barra lateral cap avall i navega endavant o enrere als costats. També es pot associar a altres ordres o skills habilitades.

El dial navega pels controls del compositor i pot modificar l’esforç de raonament. Al vídeo, aquesta navegació no convenç gaire l’autor, que preferiria poder desplaçar la pàgina verticalment. La limitació il·lustra bé la diferència entre una possibilitat tècnica i un bon hàbit: si girar el dial no estalvia més temps que usar el ratolí, deixa de tenir valor.

El dictat és la funció que més convenç

La tecla de micròfon activa el mode de prémer per parlar encara que la finestra no tingui el focus. Codex Micro no incorpora cap micròfon; utilitza el de l’ordinador. Es pot mantenir premuda la tecla mentre es parla o fer-hi una doble pulsació per gravar sense haver-la d’aguantar.

L’autor dicta una instrucció perquè Codex afegeixi a una web un botó per alternar entre mode clar i fosc. En acabar, prem la tecla de Codex per enviar el text transcrit. L’agent fa el canvi i la nova interfície funciona. És una demostració senzilla, però mostra el cas d’ús més natural del dispositiu: parlar mentre es mira una altra finestra i enviar el prompt sense tornar al teclat principal.

El dictat no és exclusiu d’aquest maquinari i també es pot activar amb altres dreceres. El valor afegit és tenir una tecla física fàcil de localitzar, juntament amb el feedback lumínic de la gravació i de l’agent. Per a instruccions llargues, l’autor preferiria activar i aturar la gravació amb pulsacions en lloc de mantenir la tecla premuda, una opció que la guia oficial sí que documenta mitjançant la doble pulsació.

Configuració i dependències

Per integrar-lo amb Codex cal obrir l’aplicació d’escriptori de ChatGPT i connectar el dispositiu. A macOS s’ha d’autoritzar Input Monitoring perquè l’aplicació rebi les pulsacions. Després d’haver-lo detectat, apareix la secció Codex Micro a la configuració, on es gestionen els xats, les ordres, el dial i la il·luminació.

Codex utilitza la primera capa del teclat. Work Louder Input permet configurar fins a cinc capes addicionals per a altres aplicacions, de manera que el maquinari no queda completament lligat a ChatGPT. Tot i així, les funcions de color en temps real, estats dels agents i controls específics depenen de la integració amb l’aplicació.

Primer veredicte: divertit, útil i molt de nínxol

El vídeo no afirma que Codex Micro faci programar millor. De fet, l’autor dubta que aporti la precisió o l’estalvi de temps que sí que ofereix una superfície física especialitzada en edició de color. El que valora és una combinació de plaer d’ús, dictat accessible i supervisió visual de diversos xats.

Per a una persona que treballa amb un sol agent i sempre té ChatGPT en primer pla, moltes funcions ja es poden executar amb dreceres de teclat o ratolí. En canvi, algú que manté diversos xats simultanis, aprova accions sovint o dicta instruccions pot aprofitar millor les tecles d’estat i els controls directes.

També cal interpretar la prova amb prudència: és el primer dia, l’autor es defineix com a novell en Codex i només crea una modificació web petita. Encara no avalua autonomia, bateria, fiabilitat del Bluetooth, ergonomia prolongada ni si les assignacions sobreviuen als canvis d’hàbit.

La conclusió més honesta és que Codex Micro no és imprescindible ni pretén ser-ho. És un perifèric car i limitat que converteix un flux de treball abstracte —agents que pensen, esperen aprovació o acaben— en llums i controls físics. Per a la majoria serà una curiositat; per als entusiastes dels comandaments personalitzats i del treball multiagent, pot ser una manera agradable de mantenir el control sense vigilar constantment la pantalla.

Contrast i context

Fonts consultades

4 fonts
  1. 01
  2. 02
  3. 03
  4. 04

Font de treball

Transcripció amb marques de temps

4 fragments
Consulta la transcripció
  1. 0:00 , obre el vídeo en una pestanya nova

    Welcome back, friends. It is July 22nd, 2026, and I just received my Codex Micro. Now, before you get pre-mad, cuz I know a lot of people are hating on this piece of hardware, just know that I love computer peripherals. I have Stream Decks, I have color grading surfaces, I have everything. I I've made custom video games for custom computer hardware that train remote head operators for camera cranes on big movies. Really expensive hardware and computers where you get to do custom things, that's kind of my thing. I love it. So, when I saw that Open AI/Work Louder was putting this project together, I bought it sight unseen. That's on site. I bought it on site as one does when they're an enthusiast. Do I expect this to change my life and make my projects massively better? No, it's not like a control surface for Resolve where like actually touching wheels actually can actually give you more finesse and time and speed. I'm not sure that this is going to do that for me, but do I think it's fun? Do I want to test it out? Yes. And for further disclosure, I am a one-trick clawed code user. I've used Codex like twice in my life, but this got me to do it again. Also, people have been asking for Codex tutorials, so here we are. Just extra reinforcement for me to actually do that.

  2. 1:36 , obre el vídeo en una pestanya nova

    So, here's it to camera. If I point it at the light, it's just incredibly reflective, but very cool packaging. Spin it around. Has branding. I mean, this box is going to go in the background over here. I just It also came with this box here, which is very nice looking. See if I can get this to auto focus. Has our little Codex logo, which the Codex app I guess rip the Codex app doesn't really exist anymore. So, we'll flip it over. Has some little strips here to open up. And this is my first time opening it. Unfortunately, the box took a little bit of a of a hit in the corner here, but I think we'll we'll survive. There it is. First time coming off. Nice box. Has a cool uh kind of like inner molding, very very Apple feeling kind of uh packaging here, really nice. Going in the back. And there it is. Uh it sort of like lifted up at me. Pretty cool. Just really nice packaging. I'm having fun with it. This is like buying a new piece of like Apple hardware. Uh very cool. So, let's let's actually take it out of the box now. So, we're going to actually take it out. Very easy. Put this here as we explore the box. Here we have a little quick start guide. Has a QR code, very nice. And it comes with a braided cable. Uh again, very very Apple feeling. And I luckily have one USB port left, USB-C port on my my little Thunderbolt dock, so we'll be able to actually try this out. Kind of purple. I didn't realize that like Kodex's color palette is purple like this. It's like this kind of purple very very light purple {slash} white. So, there's our box. Where's it going to go in the background? We'll go put it somewhere in the background for now. So, there it is in our background now. It's just very shiny. And so, here's our controller. Very clicky keyboard, I like it. So, these are kind of like um you know, like regular keyboard plastic, but these are more rubber. And they're see-through, so you can actually see the little like orange switch that that's inside of it. Here's it compared to my uh Logitech mouse, the old MX Master. So, these are the This is all you need in uh 2026 to program in and make films uh is this little keyboard. I guess your mouse sometimes, too. But again, yeah, these ones are the uh kind of plastic ones, the plastic keys. And then these are like kind of the rubber ones. And then we have a knob here that spins. Trying to show it to the the face camera for now. Has some notches on it. It's not like completely smooth. And And we have a a joystick. It doesn't really have a click, but you can kind of feel when it's like probably going up and down. So, before we do that, we have one more piece here. These are like probably the custom keys, which I didn't do a ton of research on how this works, and I'll probably not even going to get up and running today. I just wanted to unbox it. We got it. We got a clean open. Thank goodness. Okay. I didn't destroy the box. Very cool. I have a feeling like all these keys are going to explode out of here if I open it. There we go. Okay, cool. Really, really cool. Um again, we want to keep the box. These will probably just keep somewhere else. The box itself is just art. Not bad. Not bad. Okay, so we got the little keyboard box and then that's up there, so cool. So, we can look at our keys uh from above here. Maybe I can zoom in. And uh open this up. And we have a GPT-1, a terminal one maybe. A trash can, new document. And I guess if you like map them to this, you can then put them on there. Uh not sure yet. And I'm going to have to make like a fall one. I don't know what about fall's going to be on this, but we'll have some sort of fall key. And I'll just keep this on there for now. I don't feel like I'm going to be customizing this cuz I don't know how this thing works at all. And yeah, so just regular USB-C. Huh, there's like a button. I don't know if that's like a reset button. Oh, it turned on. Oh, it's wireless. You can use it wirelessly. Ah, who knew? So, that button on the back is wireless power. It came charged. So, it's already already turned on and blinking. Very fun. But anyway, here we are uh almost up and running here. We've unboxed it. We've looked at the hardware. And now I'm going to go read some tutorials and figure out how you actually get it connected, and then we'll build some project, and I'll just kind of take it for a run, you know, try not to use my keyboard at all. We'll we'll see.

  3. 5:37 , obre el vídeo en una pestanya nova

    So, here we are in ChatGPT/Codex. Uh I guess we're in Codex mode over here. Again, I'm very much a Claude code user. Uh this is like my third time in here. So, I built a little website before, but I forgot to record the audio. So, here we are again. Um so, let's do a new chat. And the default bindings right now Oh, I got to turn it on. Let's turn it on. There we go. Let it Bluetooth up. There we go. So, the default binding for the wheel, as you may have seen, is going to spin um the context between these things here. I don't find this super helpful, but maybe there's something else it can do in the future. These are going to switch between the different chats that we have. Um and then this is going to hide the sidebar. And these are just remappable, right? There's nothing like This is just kind of moving between the chats again. Up turns on plan mode, and you can just map these to what you want in the future. We have accept, don't accept, cancel, I guess. Fast mode, which I don't even know what that means in GPT, but you know, maybe that's something you do a lot. And then we have fork, which I changed this to settings because I just fork like a ton of projects. You don't want this to be fork. I think that that would be kind of a bad idea. Um so, we'll go back, and something that's important when you're getting started with this is this uh little button. Show it again really quickly is the Codex micro button, shows the battery. That's helpful. And you did have to give uh Mac OS a bunch of permissions to talk to it and whatnot, nothing crazy. We see that we have our agent keys. Um this is the knob, does composer navigation right now. And then you can click each one of these buttons or these ones rather, and you can change uh through a a lot of different bindings here that you want to go through, and maybe you can change it to other things as well. And for now, I just put mine to settings. So, for me at the moment, the only real thing that I can see using this for is the talk button here, little dirty. And uh we'll see that we have a little website that I made. I can use this to hide that. I made a little website before, but I didn't record the audio. Wasn't very difficult. Um there's no way to scroll using this at the moment. That would be cool if this scrolled the page. I think that that would be kind of helpful or like yeah, scroll this up and down. That would be something that maybe we could bind it to would be would be cool. And uh but I'll just show you quickly like the talking feature. So even with it not focused like this window isn't focused like say I'm like doing something over here or whatever. I should be able to hold this down and say the website looks great. Let's go ahead and make a dark mode version. So put a button on the top for now that allows me to switch between dark mode and light mode. And so in GPT {slash} Codex, you don't get to see the text being written as you say it. You just see the waveform and then it takes a second. So maybe it has like more accuracy in that compared to like Cloud Code. And then if we want to send this, we're going to hit the uh Codex button. So that's kind of cool. Um in Cloud Code, uh it's like Apple D or you can bind it to whatever you want obviously. A lot of the times I just click it and then talk for a long time. Uh I don't feel I definitely don't want to have to hold it down. So I'd rather be able to just like hit it and then hit it to stop it. Maybe that's it. We could set that up too. And then if we go back and bring out the agent mode, if we were going to go do something else, we can see that this is now busy and it's blue. We could go work on another project, start talking to it. And then presumably when this is done, it's going to be a different color. Okay, so the agent turned green. So we could click it and we'll see that this is now ready. I don't know Codex all that well. Like do I have to like refresh this or is it just there? Oh okay, so here's our website. We can go dark mode. Cool. Kind of a purple look. Awesome. So we made this project using our new Codex micro. Uh couldn't have done it without it of course. And uh yeah, so for now again, I see value in this chat. I mean there's probably things we can do here and seeing the status of the agents if you're going multi-agent, multi-chat like that, which I'll at the most I've ever done is three in Claude code. But you know, there's some value in that. That's kind of cool and I bet you there's some fun mappings we can do for these as well. And yeah, especially if you're just talking, you know, maybe you could get away with really just using this in your mouse, who who knows.

  4. 9:59 , obre el vídeo en una pestanya nova

    So that wraps it up for our first look at the Codex Micro. It's just a first look, first impressions. I'm a Codex noob, like I said in the beginning, I only use Claude code. Can I use this to in Claude code? Maybe maybe I'll try, but anyway, I wanted to start using Codex more just to compare. People have been asking, it's sort of my job to know Codex and Claude code. So for me, this is just one little extra reinforcement to actually use it cuz it's kind of fun. Otherwise, I'll just go to my safety, which is again Claude code. And you know, I'll report back on X and this YouTube channel if I start using it more and more and I figure out ways that I'm finding it useful. It's just day one, it's going to take a little bit to figure it out. And we'll see. Anyway, thanks for watching. Hope you like seeing this in in person. Let me know in the comments if I'm an idiot and it's a waste of money, but I I think it's fun and you know, things that make things I think are fun, they bring me joy. I'm, you know, I'm always looking for something like that. See you in the next video.