I will build an ai chatbot that answers from your documents, self hosted on dify


About this gig
Your customers and staff ask the same questions every day. I build a chatbot that answers them accurately from YOUR documents, with a citation to the source, instead of hallucinating.
What you get
- A retrieval-augmented (RAG) chatbot built on Dify, trained on your PDFs, docs, web pages or FAQs
- A source citation on every answer, and a safe "I don't know" when the answer is not in your data
- Web widget ready to embed; optional Telegram, WhatsApp or Zalo channel
- Self-hosted option: runs on your own server, so no document leaves your infrastructure and there are no per-message platform fees
- Documentation and a short handover video
Why me
Tech Lead of a live AI SaaS; I run a self-hosted Dify + n8n stack with 16 bots in production for SMEs and a public-sector agency. The boring parts are solved: chunking that retrieves, multilingual queries, rate limits, backups.
Before you order
Message me with what the bot should answer and roughly how many documents you have. I will confirm the right package or send a custom offer within a few hours.
Not included: LLM API usage (a few dollars a month for small teams; I show you how to cap it).
Get to know Hieu P.
AI Automation Engineer, n8n and RAG chatbots, full stack developer
- FromVietnam
- Member sinceSep 2026
Languages
Vietnamese, English
FAQ
Which AI model do you use?
Your choice: OpenAI, Anthropic, Gemini, or an open-source model. I recommend one based on your budget and language.
Do I need my own server?
No for Basic and Standard: I host the bot or use Dify Cloud. Premium sets up Dify on your own VPS so you own everything, including the data.
Can it answer in Vietnamese or other languages?
Yes. Multilingual retrieval and answers are supported; the Standard package covers 2 languages, and more can be added as an extra.
Who pays the AI API costs?
You do, through your own API key, so you keep control. Small teams usually pay a few dollars a month. I set spending limits so there are no surprises.
What if the bot gives a wrong answer?
It only answers from your documents and cites the source. When the answer is not in your data it says so instead of guessing, and can hand the conversation to a human.

