“What is Dify” and “how is it different from ChatGPT” are covered on the Dify product page. This article is about the questions you run into once you start building apps.
Models and cost
Q: Do I need to bring my own model API key? Who do I pay?
Yes, you bring your own, and the model usage charges are paid by you directly to the model provider (OpenAI, Anthropic and so on), not through RoamerHost. What you pay us is the hosting fee; what you pay the model provider is the usage fee. The two are separate.
The upside is that you have full control over which model you use and how much you spend, and you can switch providers whenever you like instead of waiting for the platform to support one. The cost is one extra step during initial setup, plus keeping an eye on usage and billing on the model side yourself.
Q: If I switch model providers, do I have to rebuild the apps I made?
No rebuild needed. Dify treats “the model” as a swappable setting, and the prompts, workflows and knowledge bases you build in an app are independent of any particular provider. What switching models does require is retesting quality — different models respond differently to the same set of prompts — but the structure stays as it is.
Knowledge base
Q: The knowledge base gives inaccurate answers. What can I tune?
The most common cause is a chunking strategy that does not suit your documents. RAG works by splitting documents into small chunks, retrieving the most relevant ones and handing those to the model, so how you split directly determines quality: chunks that are too small lose their context (the model gets a fragment without knowing what it is about), while chunks that are too large reduce retrieval precision (relevant and irrelevant content get mixed together).
Dify’s interface lets you adjust the chunking parameters and test retrieval results live. Take a few questions people will actually ask and iterate on those — it works better than trying to get the settings perfect in one pass.
Q: Large file uploads are failing
The plan comes with 10 GB of storage, which is generous for documents, but processing is what consumes memory — a single very large file can hit the memory ceiling while it is being chunked and vectorized. When that happens, splitting the file into several smaller ones and uploading them separately usually solves it, and the chunk quality after splitting is often better anyway.
Resources
Q: Why is the Dify plan so much more expensive than other services?
Because the architecture is different. Services like n8n are a single process; Dify is a set of services — an API service, background workers, a vector database, a relational database and a cache layer. That shows up in the specs: Dify gets 2 CPU cores and 10 GB of storage, against 1 core and 2 GB for n8n.
For the detailed architecture, see what Dify is technically good at.
Sources and further reading
Dify’s features and interface change between versions, so check the official docs before you follow any steps:
- Dify official documentation
- Dify release notes — the authority on what changed
- Dify source code and issue tracker
Publishing apps
Q: Can I embed a finished chatbot on my website?
Yes. Once an app is published it gets its own URL you can share directly, plus embed code that can render as a chat bubble in the corner of a page or as a full-page embed. Visitors do not need to log in to Dify to use it.
Q: Is there an API I can call?
Yes. Every app has its own API and key, so n8n or your own systems can call it and plug the AI capability you built into existing processes. For a worked example, see combining Dify with n8n.
Q: Which models are supported?
Mainstream providers such as OpenAI, Claude and Gemini are supported, as is any service compatible with the OpenAI API format. You can configure several providers in the same workspace and let each app pick whichever model suits it best.
Q: Which document formats does the knowledge base support?
Common formats including PDF, Word, TXT, Markdown and web pages can all be uploaded and parsed. Note that scanned documents (image-based PDFs) parse poorly — convert them to text first and retrieval quality will be much better.
Further reading
- Dify managed hosting pre-purchase FAQ: model costs, knowledge bases and commercial use
- A full introduction to Dify
- Build your first AI chatbot
- RAG knowledge bases and workflow orchestration
Want someone to build it for you?
If you would rather not put all of this together yourself, or the project is big enough that you want someone to help plan it, Roamer Tech (RoamerHost’s parent company) takes on contract work for business process automation and AI agents:
