Online AI Documentation

This topic provides an introduction to ways you can use Help+Manual's Markdown export to make your documentation available online so that it is AI-searchable.

The objective here is to provide direct AI searchable access to your documentation, so that you are sure that you are really querying your documentation and only your documentation. This is different from using Help+Manual's hybrid HTML5/Markdown WebHelp option. That does make the main AI systems aware of your documentation in time, but the systems will then also use any other information they have ingested as well and their results can and probably will include more AI hallucinations.

You can't "train" AI systems on your documentation directly

If you want to query your own documentation directly with AI, you can't train systems like Claude or ChatGPT on your documentation. (See Hybrid HTML5/Markdown WebHelp for what is possible there.) Training changes the model's underlying parameters and requires a substantial training process controlled by the model provider. It is not equivalent to uploading a set of documents and having the model permanently learn them.

Similarly, you can't just give the chatbot your entire documentation during a chat session. Except with very small projects, this would overload the context window, become extremely expensive, or both.

The solution is RAG

A RAG (Retrieval-Augmented Generation) system solves this differently. You load your documentation into the RAG system, which stores it in a special structured database. Then it acts as an agent between the user and the AI system.

Your documentation remains an external knowledge base on the RAG system. When a user asks a question, the RAG system searches the documentation for the most relevant passages and supplies those passages to the AI as context for generating the answer.

This means the AI can answer from your actual documentation without having to be trained on it, and the documentation can be updated or replaced without retraining the model. For technical documentation in particular, this is important because accuracy, traceability, and currency are more useful than trying to make the model memorize the documentation.

Online RAG systems are currently the best solution

It is possible to set up your own RAG system, but it is quite complicated particularly if you want it to make your documentation available online. The simplest solution and generally most reliable solution is to use an online RAG service. You basically subscribe to the service, upload your documentation and you're pretty much done.

Help+Manual has already done the heavy lifting for you

Help+Manual's Markdown export generates the ideal format for feeding your documentation into any kind of RAG system. You don't need a heavy, expensive pipeline that performs complex parsing of the formatted text to make it digestible for the AI system. You just need platform services that preserve your Markdown formatting, offer clean web-facing search interfaces, and natively connect to frontier models like Claude, Gemini, or ChatGPT. [1, 2, 3, 4]

The list below is not exhaustive, but it provides a good introduction to the services available. Please note that we haven't used these services ourselves, so we can't provide any guarantees. The list was provided by the AI systems themselves as representing good solutions for this scenario.

Code-Free Platforms for Customer-Facing Bots

If you want maximum simplicity or just want to give AI-searchable documentation a try, these code-free solutions are probably the first thing to try.

•Mendable.ai: This system is built specifically for technical documentation. It natively understands Markdown, respects headers for structural context, and provides customizable, production-ready chat widgets you can embed directly into your website for your own customers. [5, 6, 7, 8, 9]

•Custom GPTs via OpenAI: If you already use ChatGPT Plus or Enterprise, you can build a custom GPT. You just just upload the structured Markdown files directly into the GPT's knowledge base. It natively leverages GPT for precise, structure-aware querying. [10, 11, 12, 13, 14]

•Dify.ai: This is an open-source, fully hosted LLM app builder. You can upload your Markdown files to Dify's "Knowledge" section, choose your preferred AI model, and instantly deploy a clean, web-based chatbot UI or an API endpoint. [15, 16, 17, 18, 19]

Developer-Centric RAG Services

These platforms are more for developers who want to build their own custom UIs for their AI documentation search system.

•Vectara: This is an end-to-end RAG platform that accepts direct file uploads via console or API. Because it bypasses traditional chunking and relies on cross-attention neural re-ranking, it performs exceptionally well with pre-structured text, providing clean citations out of the box. [20, 21, 22, 23]

•Ragie.ai: A developer-first RAG-as-a-service. It allows you to upload the structured Markdown generated by Help+Manual directly via an API. Ragie handles the retrieval and sync logic seamlessly, letting you as the developer wire up the retrieved context directly to Claude or GPT endpoints in your own frontend apps. [24, 25, 26, 27]

Cloud Ecosystems (For Enterprise Users)

•Google Cloud Vertex AI Search: This is ideal if yo want high-performance Gemini integration on a larger scale. You can drop the Markdown files into a Cloud Storage bucket, link it to Vertex AI Search, and instantly get an API endpoint capable of structured, cited retrieval. [28]

[1] https://skywork.ai

[2] https://pub.dev

[3] https://www.claudepluginhub.com

[4] https://www.mindstudio.ai

[5] https://ai.gopubby.com

[6] https://repocloud.io

[7] https://mdflow.cz

[8] https://www.meilisearch.com

[9] https://abp.io

[10] https://denser.ai

[11] https://latenode.com

[12] https://simplescraper.io