Improve llms-full.txt by adding API references and adding compact version of the docs

[aglitchman]

After reading this discussion, I realised that 200K+ tokens in the current llms-full.txt file is a lot, it is not suitable for all LLMs, i.e. only Google Gemini 2.5 can handle it. In my opinion, we need to improve the documentation files for LLMs as follows (got the idea from the blog post):

• llms-full.txt: include everything about Defold, i.e. manuals + API.
• llms-small.txt: a compressed version of llms-full.txt, i.e. by removing docs about porting to consoles, IDE stylization etc, and keep docs only about programming.
• llms/docs.md: manuals only, like llms-full.txt now.
• llms/ref.md: API only.

It seems to me that this structure would be flexible enough to allow a developer to specify only documentation or APIs as the LLM context.

P.S. Actually, there is no clear instruction in the community right now about what type of content you need to ‘feed’ LLM to have high quality answers.

[aglitchman]

llms-full.txt generator is located here - https://github.com/defold/defold.github.io/blob/master/update.py#L335

[aglitchman]

llms-full.txt file is already accessable from the Internet - https://defold.com/llms-full.txt

[aglitchman]

Plus, it would be great to write a manual page or a special page for LLMs only with Protobuf formats of all Defold files.

[yerumaku]

Plus, it would be great to write a manual page or a special page for LLMs only with Protobuf formats of all Defold files.

If the template files were commented, it would be possible to use them as a basis.

[aglitchman]

I'm working on adding Reference API to the llms.txt files and want to share the current result to get feedback from the community.

ref.md - 75437 tokens 301751 characters 40005 words

Explanation. I added the Defold API, the Lua API and all the official extensions. Without C, it's a huge piece of text, which is not needed for Lua development and should probably be put in a separate file. To reduce the size, all functions are presented as Lua code and a short description, without a detailed explanation of all parameters and return values. But all examples are included. Maybe this will be enough for AI to write working code. You should check it out!