Trongate docs as LLM-friendly Markdown - General Discussion

Trongate docs as LLM-friendly Markdown
5 months ago

#1

Hi everyone,

I'd like to share a new community resource that I've been working on: Trongate Docs (Markdown) - a complete conversion of the Trongate v1 HTML documentation into Markdown format.

https://github.com/DaFa66/trongate-docs-md

This project takes all of the Trongate v1 HTML documentation from DC's GitHub repo and converts it into Markdown files, making the docs more accessible for:
• AI/LLM Integration: Perfect for RAG systems, vector databases, and AI coding assistants
• MCP Servers: Use with Context 7 or custom Model Context Protocol implementations
• Static Site Generators: Deploy with MkDocs, Docusaurus, VuePress, GitHub Pages, etc.
• Offline Development: Access documentation without an internet connection

All sections are converted:
• php_framework/ - Core PHP framework documentation
• reference/ - API reference and technical specifications
• trongate_css/ - CSS framework and styling documentation
• trongate_mx/ - Trongate MX JavaScript library docs

useful for:
• Feeding Trongate documentation to AI coding assistants for better context
• Creating offline reference materials
• It has been uploaded to Context 7 with some good results
     - https://context7.com/dafa66/trongate-docs-md
• Also handy for tools like `AnythingLLM` [https://anythingllm.com/] & `LM Studio` [https://lmstudio.ai/]
     - for importing into a vector database like LanceDB or Pinecone
     - local AI, on your computer
     - API calls to integrate into your Trongate app [i.e. chatbot]

Cheers! DaFa

5 months ago

#2

This is actually a super useful thing, Simon. Thanks for this.

When v2 comes out, I really want all those AI bots to know and understand the framework (Trongate v2). So, the plan is to remove a massive bunch of content including YouTube videos and even the v1 docs. All of that material, if left lying around the web, could cause confusion.

Nevertheless, I think the idea of having the v1 docs somewhere where somebody can access them might be a good idea. Perhaps we can arrange a meeting at some point to figure out a way of keeping v1 docs available.

Just a thought.

3 months ago

#3

Thanks for the context7 docs, useful.

I have built my own LLM docs for toying with V2 which i'd like to share in hope that others find it useful.
The reason i'm posting this as a reply and not a thread is i thought Dafa could have a look and update his docs from my docs/agents.md :)

llm.txt Entrypoint, most LLMs pick up on this right away.

docs/agents.md this is the most important file, steers code style and patterns for coding agents

docs/system-prompt.md not necessary, but helps set the stage for the coding assistant to produce better quality code

docs/project.md optional but recommended, this ensures the coding assistant retains the project features and domain in context.

Adapt the files to your liking and let the coding assistant build.

The prompt can be as simple as:

P.S. if you're using Claude Code i can recommend installing Ralph Wiggum, PHP-LSP & Brainstorming plugins.

3 months ago

#4

Thanks for that, was thinking the same. Also looking into adding skills but have to keep a check on the context window of the LLM used, and try keep it in the 'smart zone' of < 40%, anything over 60% and the model dumbs down, 80%+ and you get into 'hallucination zone'.

3 months ago

#5

Glad to hear.
I’m not deep into skills, to me it looks like shortcuts in same vein as stored procedures in SQL. :)

Mhm and also being very aware that bigger token limit does not necessarily mean it can “comprehend “ bigger prompts or code bases.
It’s important to rephrase short, concise and to the point.
Occam’s razor style.

2 months ago

#6

Cheers sasin91 for the headstart. Been scrapping the V2 docs as pdf to feed NotebookLM.