Customers keep asking the same questions. Your docs already have the answers.
The hard part is getting those answers into a chat that sounds helpful, stays on your material, and does not invent policy. You want to show stakeholders something real before you bolt on ticketing, auth, and retention rules.
AI Support Bot is a full-stack demo you run yourself: load your knowledge, chat in the browser, and see how retrieval-backed replies behave.
Your docs in the chat window, not a generic model monologue.
What you get
- A working chat UI wired to your loaded knowledge. Answers are meant to draw on material you provide, not free-form guessing alone.
- Room to refresh what the bot knows. The project supports dynamic dataset loading so you can update sources without rebuilding the whole app from scratch.
- A quality lens on replies. Responses are evaluated on relevance, accuracy, coherence, completeness, creativity, tone, and alignment with intent, so you can talk about quality with stakeholders, not just vibes.
- One-command local spin-up. Docker Compose builds and runs the stack; you tune secrets in the backend environment file when needed.
- Honest limits up front. Chat history is not stored in a database yet, and wrong answers remain possible, so you still plan human escalation for high-stakes topics.
Picture your Tuesday
You are prototyping an internal help bot (an illustrative case). You load a slice of your handbook, open the web chat, and ask the questions your team actually types in Slack. You watch whether answers stay tied to what you uploaded, then share the screen in a roadmap meeting instead of a slide full of architecture boxes.
Keep exploring
- AI Customer Feedback Analyzer if you need themes from messy comments before you write the docs the bot should cite.
- Awesome LLM Experiments for focused notebooks that walk through RAG patterns you can compare against this app.
Ready to try it?
- See it on GitHub for the README, flowchart, and Docker Compose setup
- Price on this site: Free, licence AGPL-3.0









