Blog
Notes on choosing, deploying and operating open-source AI support.
Chatwoot Captain or bolt-on Dify? Choosing your AI brain
Chatwoot's built-in Captain can reply with an LLM directly — but it trades off against a bolted-on Dify Agent Bot. Clear criteria here.
10 common pitfalls when self-hosting open-source AI support (with fixes)
The 10 most frequent issues we've seen rolling out Chatwoot + Dify / FastGPT over the past year — repro and fix for each.
Tuning FastGPT KBs — from 0.62 to 0.89 accuracy, step by step
FastGPT defaults give ~60% retrieval accuracy. Tuned right it reaches ~90%. Step-by-step measurements for chunking, augmentation, reranking.
Fully local AI support — Open WebUI + Ollama + Qwen on one GPU box
Under strict no-data-egress mandates, can one GPU server run the entire AI support stack? Hardware, models, KB, exposure — end to end.
n8n as central orchestration for AI support — 10 real automations
The business-logic layer of an AI support stack gets ignored. n8n's 800+ nodes glue Chatwoot, Dify, CRM, ERP and Slack together.
RAGFlow vs Dify KBs — head-to-head on one complex PDF
A 30-page product manual with tables, two-column layout and scanned pages. Indexed with both. Top-5 accuracy and answer quality compared.
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.
Dify vs FastGPT — 8 dimensions to decide
The two most popular open-source LLM app platforms in China overlap heavily. This compares them across 8 dimensions and gives recommendations.
KB preparation in practice — from Notion / Lark / Confluence to RAG
A bad KB ruins any LLM budget. End-to-end repeatable pipeline — export, clean, chunk, augment, evaluate — with Python.
How to evaluate your RAG quality — 5 metrics, 3 toolkits
"Feels better than last week" is not measurement. Five quantitative RAG metrics and three open-source evaluators.
Production-grade Cloudflare Tunnel tuning for AI support
One-command Cloudflare Tunnel for Chatwoot + Dify hides production gotchas — WebSocket drops, timeouts, monitoring blind spots. A checklist.
2026 Chinese embedding benchmark — bge-m3, Conan, m3e, bce, OpenAI
Same Chinese KB, same real support questions — five embedding models compared on retrieval accuracy, speed and cost.
Chatwoot deep review — 6 months in production, the unvarnished take
We ran Chatwoot 4.x in production for 6 months — 40k+ tickets, 12 agents, full debrief on Inbox, Captain, API and gotchas.
Dify workflow patterns — 10 we actually use in production
Ten Dify workflow patterns that survive production — KB fallback, context compression, concurrent tool calls, retries and more.
Why self-hosting an AI support stack pays off in 2026
SaaS support keeps getting pricier while open-source matures. A framework across cost, compliance and control.
Migrating from Zendesk to Chatwoot — translating tickets and triggers
Zendesk's ticketing is heavier than Intercom's; the hardest part of migrating to Chatwoot is rewriting Triggers / Automations / Macros.
Migrating from OpenAI GPTs to Dify — bring your Custom GPT home
Custom GPTs are great for tinkering but limit you when you go commercial, need an API or want to swap models. Step-by-step Builder → Dify migration.
Migrating from Intercom to Chatwoot — the complete path
Data export, field mapping, parallel running, cutover — a battle-tested playbook with scripts and traps.
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 · A B2B SaaS migrated from Intercom to Chatwoot in 6 weeks
A 2,000-customer, $2M ARR B2B SaaS completed Intercom → Chatwoot migration in 6 weeks. Full timeline, headcount, scars.
2026 Chinese-support LLM bake-off — Qwen, DeepSeek, GLM, Doubao, ERNIE
Same support prompt and knowledge base — which of the five China-trained LLMs ships the best AI support? 200 real questions decide.
Case · A securities firm shipped a fully local AI support — the compliance path
No data egress, regulator approval, internal + external audit, DR plan. The real compliance journey for a mid-sized securities brokerage's local AI support.
Case · An online-edu promo day with 50× traffic — engineering notes
A coding-education company handled 120k conversations in 24 hours during 618 sale, AI deflection 81%. How engineering pre-tested, rate-limited, degraded gracefully.
Case · How a healthcare provider built RAG with zero PHI leakage
An internet hospital launched AI health consultations. The hardest part — PHI must never reach the LLM. Their 3-tier KB + Pipeline redaction + audit architecture.
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.