Open-source components · On your own server
Let AI answer most of your support questions
SupportBot greets your customers through a chat widget on your site; it answers from your documents, hands unresolved conversations to a human agent, and reports every week on what it should learn next.
Features
On top of Chatwoot + Dify, the layer your support team actually uses.
Answers from your knowledge base (RAG)
The bot doesn't make things up: it answers from the documents you upload and cites which document each answer came from.
Human handoff
Conversations the bot can't resolve are handed to a human agent in the same window; the customer loses nothing.
Weekly AI report
Every Monday: topic distribution, satisfaction, knowledge gaps, and concrete improvement suggestions.
Learning loop
Questions the bot couldn't answer land in the report; upload the document and it answers next time. Your feedback shapes the next report.
A widget that matches your brand
Colors, logo, welcome message, and suggested questions are managed from the panel; accessibility contrast guarding is built in.
MCP + API
Claude and other MCP-capable clients operate the platform through 44 tools: upload documents, trigger reports, run test scenarios…
Your data stays yours
Runs on a single VPS with Docker Compose; conversations and the knowledge base live in your own PostgreSQL.
Multi-model LLM
Anthropic, OpenAI, Google, Groq, DeepSeek, Mistral, xAI… Pick the provider in the panel and verify with one click.
How does it work?
Go live in three steps.
- 1
Install
Fill in infra/.env and run docker compose up. Zero-touch provisioning sets up the accounts, the bot, and the knowledge base by itself.
- 2
Upload your documents
Drop your FAQs, guides, and policies into Panel → Knowledge Center; the bot starts answering from them within minutes.
- 3
Paste the widget
Add the one-line snippet from Panel → Integrations to your site. That's it — the bot is talking to your customers.
Whatever format your knowledge lives in
Documentation
A bilingual user guide, from setup to MCP integration.
Frequently Asked Questions
Where is my data kept?
Entirely on your own server: conversations, the knowledge base, and configuration live in your own PostgreSQL database. The only outbound traffic is the model calls to the LLM provider you choose.
Which AI models can I use?
Anthropic (Claude), OpenAI, Google, Groq, DeepSeek, Mistral, xAI, and more. Provider and model are selected in the panel; your API key is stored encrypted.
What happens when the bot answers incorrectly?
Every answer is logged with its source documents. Feedback you leave on a conversation feeds the weekly report; questions the bot couldn't answer are collected as 'knowledge gaps' — add the document and the gap closes.
What do I need to install it?
A Linux VPS with Docker Compose and a domain name. Installation is a single command; accounts, the bot, and the knowledge base are set up automatically (~10 minutes).
How do I connect WhatsApp and Instagram?
Through Chatwoot in a few steps: enter your WhatsApp Cloud API credentials, and define a Meta app identity for Instagram. New channels bind to the bot automatically.