Using AI with ProcessWire: Where Do I Even Start?

[zilli]

Hey @ryan and fellow AI enthusiasts,

This might be slightly off-topic, but I think it could be really valuable for the ProcessWire community if someone created a practical guide or tutorial on building projects with PW and AI tools.

I’m new to this space, and honestly, it feels a bit overwhelming because there are so many different approaches and workflows.

For example, after installing Codex or Claude, what’s the next step? Should I create specific .md files for context or instructions? If so, which ones? Is it better to have the AI read the entire ProcessWire codebase first, or should I simply ask it to build a module or website directly?

I’d really appreciate any guidance, workflow examples, best practices, or general tips from people already using AI successfully with ProcessWire.

[Peter Knight]

Hi Zilli
PromptWire is available, and it handles 100% of my AI development in ProcessWire.
It's built as an MCP from Cursor to AI, and it's free.
https://www.peterknight.digital/docs/promptwire/
It can build almost anything with the right prompts, and it's really quite powerful. 

Any questions, just DM or reply here. 

Cheers
Peter

 

[szabesz]

20 hours ago, zilli said:

Claude

Hello, I have not yet used Codex, but I've been using Claude Code for three weeks, but not yet for ProcessWire projects, but I will surely use it for ProcessWire projects too.

I recommend watching some good tutorials first, like: https://youtu.be/uogzSxOw4LU

Next, setup the MCP servers you need, I recommend https://context7.com/ and @Peter Knight's PromptWire, and any other MCP servers Context 7 cannot help with.

For connecting to systems, I recommend composio.dev

Also, do not forget: https://processwire.com/modules/agent-tools/

After you have setup Claude Code in Cursor (for example), just ask Claude whenever you need info on anything.

And make full backups regularly, AI can screw things up fast... ;)

 

 

Edited yesterday at 02:21 PM by szabesz
typos

[ryan]

@zilli I'm kind of just getting started too, as a couple of months ago I knew nothing about Claude Code or Codex or any of the other tools. I also felt overwhelmed which led me to avoid the AI stuff for awhile. Too much noise. But my client started using Claude Code and I started getting interested in it. Start with Claude Code or Codex, they are basically identical. Codex is a better value overall IMO, but Claude might be slightly easier to start with.

To get started, install Claude or Codex (use instructions on their site). If using Claude, and you have a choice of "model" choose "Sonnet". Claude's premium model is "Opus" but they don't give you enough usage of it on the $20/month plan. Whereas Sonnet is still very smart and lasts for quite awhile on the $20/month plan. If using Codex, you'll get GPT 5.5 which is their latest model  and you get plenty of usage on the $20 plan. That's why I say Codex is a better value. 

The best way to get started is just start asking Claude or Codex questions about how to do things. Create a hello-world/ directory somewhere on your localhost. If you are comfortable with the terminal, go to the directory and type "claude" (or "codex"). It will ask you to login to your Claude or OpenAI account. Then you are ready to go. If not comfortable with the terminal, then stick to the Claude or Codex desktop apps, which can do just as much and maybe more. Ask Claude or Codex to do something like "create a hello-world.html file" just to get a feel for how it works. Ask it to continue making edits to it just for fun. Once you feel comfortable with this, then you are ready to use it in ProcessWire. 

Quote

Should I create specific .md files for context or instructions? If so, which ones? 

No need to start creating any .md files at first. Once you get going, ask Claude when/if you should create a CLAUDE.md file (or AGENTS.md for Codex and others, same thing). These files are really only useful once you are working on an actual project and want to provide instructions in a file that gets read every time, rather than pasting them into a prompt. But technically there's very little difference between providing something in an .md file and just providing it in a prompt, at least until it comes time to /compact or /clear (but you can learn about that later). 

Quote

Is it better to have the AI read the entire ProcessWire codebase first, or should I simply ask it to build a module or website directly?

Both Claude and Codex are already well trained on ProcessWire, and they don't need to read the codebase unless they want to. But I do think it's beneficial to run the latest dev branch version of ProcessWire, which includes documentation for AI agents built in. Installing the AgentTools module will be worthwhile for Claude or Codex to be able to most effectively use ProcessWire's API and use built-in tools for pulling documentation, site/schema maps and booting ProcessWire on its own.

You don't need to configure any agents in AgentTools for this kind of usage. Just install the module and tell Claude/Codex to go read the /site/modules/AgentTools/AGENTS.md file, and it will be ready to do anything in ProcessWire. If you later want to add agents in AgentTools, that will let you use Claude/Codex inside of the ProcessWire admin for making edits, creating migrations, etc. But you don't need that to get started. 

I haven't yet even tried any tools other than just plan Claude and Codex (and AgentTools)--so far I haven't felt like I need anything more. I'm also not using any MCP servers. The others here can better speak to those tools and such. Every morning while making and eating breakfast, I do watch YouTube videos on using Claude and Codex though. It's become part of my boot up sequence. 🙂
 

[zilli]

Thanks for the write-up, @ryan.

[diogo]

If you want to try cheaper models, while the same experience as Claude Code, check out Deepseek TUI, or Openecode or Aider for multi model.