I mean docs are largely written for an LLM-in-a-harness. That’s how it goes! If the LLM bootstraps with the right understanding of the universe and knows how to quickly build specific context flavors… life is good.
I like this a lot. What I would love to see is a panel to run management commands and see their output. Would be great in services like Google Cloud Run where you cannot access a shell anymore like you could on Heroku.
Looks great -- always wished the admin panel came with more configurable bells and whistles. I've been exploring Quarkus recently (https://quarkus.io/), and it has a Dev UI with a similar extensible "panels" pattern. It's a bit different than Django since it's not for running in prod, but nonetheless it's pretty helpful.
sort of a tangent, but quarkus also has a concept of "dev services" that are monitorable via the dev UI. It uses Testcontainers to start and autowire runtime deps (postgres, redis, keycloak, etc.). Pretty pleasant experience to get the whole stack spun up and observable alongside the dev server.
I like the way each panel is its own separate package on PyPI and the system picks them up via setuptools entry points. It's a neat implementation of a plugin pattern.
Django Admin definitely needs extensions like this. I hope someday they make it a stronger more capable Admin UI. Their own docs if I remember correctly tell you to build your own UI if you're hitting limits with the admin UI itself, which is fine, but there's so much OOTB that works nicely for the admin UI.
I like the spirit of this, and could see Django heavy shops wanting to add bits and pieces that display tooling / services they care about in Django admin.
I think its good advice to avoid the admin for customer facing use cases. But for internal facing tools It seems pretty wasteful to not use the built in admin - it has all the bells as whistles to build upon (auth, permissions, etc.)
I love the sentiment and ambition in this! The Django admin is a core reason why I still choose Django over other solutions. I tell my team that the Django admin CRUD is our backstop when we encounter issues in our frontend UI. Thank you for tooling it out more!
I think any large enough django project has toyed around with extending the admin in some way. Hopefully this project can help establish a standard to make this sort of thing easier.
i like it, but I think i would rather have a proxy, or atleast an auth redirect to those different tools.
I used to have flower at myapp.com/flower using an auth redirect in nginx to a simple view in django that made sure it was an admin user. I think if you can make that setup easier to leverage existing tools that would be nicer than rebuilding everything.
Totally understand - I am a long time flower user for example, and I am familiar with having to harden that installation a bit.
What I'm aiming for here is slightly different - keeping everything inside Django so there are no extra services to run or configure or proxy. As long as you surface the admin somewhere, then that is the place to find your tooling (including celery monitoring)
There will always be room for both approaches. A lightweight proxy/redirect could be something to explore in the future.
I love this idea. I see the AI era having 2 competing views when building something new:
1. Build X with pure <language of choice>. Why? LLMs will have less context needed, and onboarding engineers would be easier since there’ll be less overhead and opinionated frameworks knowledge required
2. Build X using well establish frameworks. Painful in the beginning since you’ll not only need language knowledge, but framework knowledge. The upshot, is scaling and maintainability
I love that this ecosystem will heavily pressure teams to consider (2) more and more — solving the very real “AI slop” problem
In my view. Building things with AI creates the need for common patterns and guardrails (i.e. frameworks) Then as these new apps become productionalized - tooling that fits your framework starts to become more important.
In that sense, AI increases the need for good patterns around observability. This project aims to make this a little easier to do for Django right from inside the framework as opposed to an external service.
I think even if AI handles more of the CRUD side, you still need to understand what’s happening in the system once it’s running - this is where this project fits in.
To your point about framework use because of AI: As more applications are being built because of lowering barriers, I think it makes sense for full stack monolithic frameworks to be used more frequently.
>I think it makes sense for full stack monolithic frameworks to be used more frequently.
Why? I believe full stack frameworks solved a problem for human coders, not AI coders. In fact they are only a limitation for AI going between programming language and runtime.
I mean for one thing your garden variety LLM had been substantially trained to handle Django. That is less context for it to bootstrap every time you summon it.
Just like rolling your shitty homebrew framework is a bad idea because only you understand it, the same is probably true with LLMs. Sure they’ll scan the bejesus out of your codebase every time they need to make a change and probably figure it out eventually… but that is just a poor use of limited context. With something mainstream, the LLM already has a lot about the universe in its training. Not to mention an ecosystem of plugins, skills, mcp servers, wizbango-hashers, and claberdashers. All there for the LLM to use instead of wasting tons of time, tokens and money perpetually relearning your oddball, one-off, rat infested homebrew framework.
There is much more engineering and testing (and probably AI training) in python and a web browser than there is in django. same with EG bash and linux vs ansible. that is what I mean by 2010s era frameworks - JSON/YAML easy wrappers with opinionated defaults and consistent interfaces.
AI has no problem going from programming language -> runtime without human-convenient middleware. So I am NOT implying to create your own django on the way to creating your CRUD app. I think you can make a CRUD app based by listing all the features you want. Including, if you really want an in-band administration feature like phpmyadmin or django admin, you could have AI generate something that pipes any system command to the web app.
Suit yourself really. maybe there's more training data for CRUD apps in python than C, but I don't think it's too hard to implement the fundamentals of a web app in any language if you're also using a web server.
Most webapps aren't that popular therefore don't use that much computation anyways, so theres a point of diminishing returns on making your CRUD as efficient as scientifically possible. some prefer a managed runtime so that a bug causes EG python to crash instead of the consequences of a bug in native code, but that can be mitigated easily enough as well.
But LLM will figure it out so why not take free speed?
BTW, also if we're getting rid of a web framework and letting the LLM write specialized code for the various CRUD operations, why not also get rid of Postgres/MySQL/Redis and let LLM also write specialized code for reading, writing, and querying the various business objects?
I think that explains some of the value for this project a bit better
I like the idea it can help for initial inspection and smell detection
README and site were definitely optimized for speed over perfection. The panels themselves got a bit more attention.
Curious what you’d want to see improved on the docs/site side.
sort of a tangent, but quarkus also has a concept of "dev services" that are monitorable via the dev UI. It uses Testcontainers to start and autowire runtime deps (postgres, redis, keycloak, etc.). Pretty pleasant experience to get the whole stack spun up and observable alongside the dev server.
I like the spirit of this, and could see Django heavy shops wanting to add bits and pieces that display tooling / services they care about in Django admin.
I think its good advice to avoid the admin for customer facing use cases. But for internal facing tools It seems pretty wasteful to not use the built in admin - it has all the bells as whistles to build upon (auth, permissions, etc.)
I used to have flower at myapp.com/flower using an auth redirect in nginx to a simple view in django that made sure it was an admin user. I think if you can make that setup easier to leverage existing tools that would be nicer than rebuilding everything.
What I'm aiming for here is slightly different - keeping everything inside Django so there are no extra services to run or configure or proxy. As long as you surface the admin somewhere, then that is the place to find your tooling (including celery monitoring)
There will always be room for both approaches. A lightweight proxy/redirect could be something to explore in the future.
1. Build X with pure <language of choice>. Why? LLMs will have less context needed, and onboarding engineers would be easier since there’ll be less overhead and opinionated frameworks knowledge required
2. Build X using well establish frameworks. Painful in the beginning since you’ll not only need language knowledge, but framework knowledge. The upshot, is scaling and maintainability
I love that this ecosystem will heavily pressure teams to consider (2) more and more — solving the very real “AI slop” problem
In my view. Building things with AI creates the need for common patterns and guardrails (i.e. frameworks) Then as these new apps become productionalized - tooling that fits your framework starts to become more important.
In that sense, AI increases the need for good patterns around observability. This project aims to make this a little easier to do for Django right from inside the framework as opposed to an external service.
I think even if AI handles more of the CRUD side, you still need to understand what’s happening in the system once it’s running - this is where this project fits in.
To your point about framework use because of AI: As more applications are being built because of lowering barriers, I think it makes sense for full stack monolithic frameworks to be used more frequently.
Why? I believe full stack frameworks solved a problem for human coders, not AI coders. In fact they are only a limitation for AI going between programming language and runtime.
Does browsing this website not qualify?
Just like rolling your shitty homebrew framework is a bad idea because only you understand it, the same is probably true with LLMs. Sure they’ll scan the bejesus out of your codebase every time they need to make a change and probably figure it out eventually… but that is just a poor use of limited context. With something mainstream, the LLM already has a lot about the universe in its training. Not to mention an ecosystem of plugins, skills, mcp servers, wizbango-hashers, and claberdashers. All there for the LLM to use instead of wasting tons of time, tokens and money perpetually relearning your oddball, one-off, rat infested homebrew framework.
Nothing has changed really…
But LLM will figure it out so why not take free speed?
BTW, also if we're getting rid of a web framework and letting the LLM write specialized code for the various CRUD operations, why not also get rid of Postgres/MySQL/Redis and let LLM also write specialized code for reading, writing, and querying the various business objects?