Build an AI customer support system that actually works
Curated open-source projects, proven combos and step-by-step deploy guides — so you can ship a smart support stack without overspending.
Don't know where to start? Three picks point you to the right path
Answer three questions — we point you to a stack, a playbook, and a deploy guide.
Featured combos
Chatwoot + Dify — the classic AI support combo
Chatwoot as the omnichannel inbox and human-in-the-loop surface; Dify as the AI brain and knowledge base. The safest starter combo for SMB teams.
Dify + RAGFlow — two-layer stack for complex documents
Dify owns the app layer and workflow; RAGFlow handles retrieval over tables, scans and complex layouts that defeat simple chunking.
n8n + Chatwoot + Dify — business-logic-driven support hub
When AI support needs to call orders / refunds / CRM / email across many systems, n8n becomes the orchestration hub that pulls all business logic into one place.
Open-source tools
Chatwoot
MediumAn open-source, omnichannel customer engagement platform — an Intercom alternative
Dify
MediumOpen-source LLM app platform — visually orchestrate agents and workflows
AnythingLLM
EasyAll-in-one local knowledge base and RAG application
Botpress
MediumDeveloper-focused visual platform for conversational bots
Erxes
HardOpen-source customer experience platform — CRM + marketing + support in one
Coze Studio
MediumByteDance's open-source AI Agent development platform
FAQ
How do open-source AI support stacks compare to SaaS like Intercom?
You own the data, you can deeply customize prompts, knowledge and flows, and cost is lower past ~5000 monthly conversations. SaaS wins on out-of-the-box experience and vendor support.
What is the minimum cost to ship a full AI support stack?
Chatwoot + Dify on a 4 vCPU / 8 GB VPS ($20-$40/mo) plus LLM tokens — typically $40-$120/mo to start, excluding labor.
Can a team without ML engineers self-host?
Yes. Chatwoot, Dify, AnythingLLM, Typebot and LibreChat all ship docker-compose installs. Only Rasa requires real Python engineering.
Which models and embeddings work best for Chinese?
Qwen / DeepSeek / ERNIE via OpenAI-compatible APIs for general chat; Qwen2.5-7B/14B for local inference; bge-m3 for embeddings.
Recent posts
AI support KPI framework — how to prove the AI actually saved money
Don't measure deflection alone — here's a 4-layer 12-metric framework covering experience, cost and training.
Case · A cross-border e-commerce running 80k daily orders — 6 months on Chatwoot + Dify
A Shopify + WhatsApp e-commerce shop migrated from Intercom to Chatwoot + Dify. Six months of real numbers — cost, deflection, satisfaction.
Case · An open-source project handled a 300k Discord community with Chatwoot + Dify
A 18k-star OSS project went from drowning in Discord to clean AI + volunteer mod tiers handling 300k community asks.