Inspiration

I always was into widgets since my first Android G1 Phone, I remember that classmates were jealous that with a glance on my phone I already knew where is the next lesson. Now I feel that the widgets are still very underrated and poorly treated by application authors or sometimes event not added at all even when app functionality screams to have it's own widget. The other thing was Notifications Fatigue and Digital Minimalism. Every app throws at us ton of notifications, now there are agents which can send notifications as well, or what's even worse they are writing to us on many different channels or one channel - if you do not read it instantly you probably forget about it or didn't notice. And that's the problem I'm resolving.

What it does

The solution consists of two parts: an iOS App and companion backend server which is mainly for Agents reachable via MCP or REST protocol. It let's your Agent to publish informative cards on an auto-rotating widget which is a part of application. You just add new named board, link this with added on home screen widget, and copy invitation message for Agent. Then agents is able to see your configured boards, can be asked to report any thing you want them to, but in a non intrusive way.

How we built it

I talked this thoroughly with GPT Voice, was consulting if such idea for an app is even doable and what is needed. Then we planned functionalities, types of cards (native swift components card - least flexible but native and fast, image where you can put any http accessible image to the widget surface, and the most flexible HTML build widget which then renders to Image as iOS limits the possibility of rendering html on widgets. So with very good specification and setting up manually Apple Push Notifications so the server can notify a phone to wake up and do something, the first version was almost one-shotted. But then there were new concepts tested, added Discover section that future user could maybe publish more general cards that someone could subscribe on their screen - I think there is some kind of potential here even for companies maybe to use it as constant showing gamification leader board for sales mans.

Challenges we ran into

The main challenge was to teach Agents the rules of using efficiently the space of the widget not making any visual elements go under the widget notch which is used for prev/next selection, shows card number, pauses rotation and let's you discard a card from widget itself. And solution to that blew my mind with GPT 5.6 Sol and Codex with computer use. Codex set up Openclaw instance, authorized and activated MCP on it, created different Agents with different models (all GPT 5.6 family, GLM and Minimax) and via MCP talked with those agents about what it would like to Agent post on the widget, then with computer use validated if it is displaying ok, if not told where is a problem and request for fix until this was ok. Codex worked autonomously for 10 hours using over 10mln tokens just on GPT not mentioning the Openclaw tokens used. And this was a huge help, to make little tweaks in specs to make any Agent with any Model prepare proper looking informative card.

Accomplishments that we're proud of

I'm really proud of the HotSpot zones that can be added during widget defining by Agent. I almost forgot about this functionality that we developed it and then I had an a-ha moment. I developed this with Codex mainly via transcriptions on Codex Mobile App during dog walks with my elderly twins and because of this I was not able to test new app version if the change needed app rebuild. After getting home I always typed Codex to put new version on my Iphone, so one time I thought what if I could press a button and have this installed or maybe even widget?! ;) So I asked Openclaw to add new card with HotSpot on it which calls an webhook to my macmini to install the app, and it worked seamlessly - that really amazed me and opened new possibilities to be using the widget for (another test was random xkcd comic refreshed with webhook on the widget).

What we learned

I'm glad that I snapped and finally gave Codex and GPT Sol access to computer use, this was a brake-trough have easily and smoothly it handled the simulator even with adding and configuring the widget on simulator by itself. I explored the options for the OpenSource'ing the project, and even prepared it for self-hosted version that when you login you can change the default server to your own.

What's next for AgentsBoard

I really think about opensourcing it or maybe to offer this for Openclaw Foundation to adopt this on their mobile app if they would be interested? For sure there will be an Android version but this was too complicated for this hackathon. I'm preparing a business use cases for it so maybe it cloud also have some small fee for companies to use - this is something new to me and for now I'm investing my own money for Apple License, servers and Codex Plan. The worse scenario is that the App will be Opensourced and used only by people like me that care of digital calmness who want to stay informed without interrupting notifications.

p.s. Repo is called openboard cause this I wanted as a name but Apple said it's restricted that's why I finally chosen AgentsBoard

Share this project:

Updates