<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>flag92 — Blog</title><description>Reviews, practice, migrations and thinking on open-source AI customer support.</description><link>https://flag92.com/</link><language>en-US</language><item><title>Chatwoot Captain or bolt-on Dify? Choosing your AI brain</title><link>https://flag92.com/en/blog/chatwoot-captain-vs-dify/</link><guid isPermaLink="true">https://flag92.com/en/blog/chatwoot-captain-vs-dify/</guid><description>Chatwoot&apos;s built-in Captain can reply with an LLM directly — but it trades off against a bolted-on Dify Agent Bot. Clear criteria here.</description><pubDate>Wed, 25 Mar 2026 00:00:00 GMT</pubDate><category>selection</category><category>Chatwoot</category><category>Dify</category></item><item><title>10 common pitfalls when self-hosting open-source AI support (with fixes)</title><link>https://flag92.com/en/blog/top-10-pitfalls/</link><guid isPermaLink="true">https://flag92.com/en/blog/top-10-pitfalls/</guid><description>The 10 most frequent issues we&apos;ve seen rolling out Chatwoot + Dify / FastGPT over the past year — repro and fix for each.</description><pubDate>Wed, 18 Mar 2026 00:00:00 GMT</pubDate><category>pitfalls</category><category>practice</category></item><item><title>Tuning FastGPT KBs — from 0.62 to 0.89 accuracy, step by step</title><link>https://flag92.com/en/blog/fastgpt-kb-tuning/</link><guid isPermaLink="true">https://flag92.com/en/blog/fastgpt-kb-tuning/</guid><description>FastGPT defaults give ~60% retrieval accuracy. Tuned right it reaches ~90%. Step-by-step measurements for chunking, augmentation, reranking.</description><pubDate>Wed, 25 Feb 2026 00:00:00 GMT</pubDate><category>deep-dive</category><category>FastGPT</category><category>RAG</category></item><item><title>Fully local AI support — Open WebUI + Ollama + Qwen on one GPU box</title><link>https://flag92.com/en/blog/all-local-ai-support/</link><guid isPermaLink="true">https://flag92.com/en/blog/all-local-ai-support/</guid><description>Under strict no-data-egress mandates, can one GPU server run the entire AI support stack? Hardware, models, KB, exposure — end to end.</description><pubDate>Wed, 18 Feb 2026 00:00:00 GMT</pubDate><category>deep-dive</category><category>local inference</category><category>Open WebUI</category><category>Ollama</category></item><item><title>n8n as central orchestration for AI support — 10 real automations</title><link>https://flag92.com/en/blog/n8n-orchestration-for-ai-support/</link><guid isPermaLink="true">https://flag92.com/en/blog/n8n-orchestration-for-ai-support/</guid><description>The business-logic layer of an AI support stack gets ignored. n8n&apos;s 800+ nodes glue Chatwoot, Dify, CRM, ERP and Slack together.</description><pubDate>Wed, 11 Feb 2026 00:00:00 GMT</pubDate><category>deep-dive</category><category>n8n</category><category>automation</category></item><item><title>RAGFlow vs Dify KBs — head-to-head on one complex PDF</title><link>https://flag92.com/en/blog/ragflow-vs-dify-knowledge/</link><guid isPermaLink="true">https://flag92.com/en/blog/ragflow-vs-dify-knowledge/</guid><description>A 30-page product manual with tables, two-column layout and scanned pages. Indexed with both. Top-5 accuracy and answer quality compared.</description><pubDate>Wed, 04 Feb 2026 00:00:00 GMT</pubDate><category>deep-dive</category><category>RAGFlow</category><category>Dify</category><category>comparison</category></item><item><title>AI support KPI framework — how to prove the AI actually saved money</title><link>https://flag92.com/en/blog/ai-support-kpi-framework/</link><guid isPermaLink="true">https://flag92.com/en/blog/ai-support-kpi-framework/</guid><description>Don&apos;t measure deflection alone — here&apos;s a 4-layer 12-metric framework covering experience, cost and training.</description><pubDate>Wed, 29 Apr 2026 00:00:00 GMT</pubDate><category>ops</category><category>KPI</category><category>metrics</category></item><item><title>Dify vs FastGPT — 8 dimensions to decide</title><link>https://flag92.com/en/blog/dify-vs-fastgpt/</link><guid isPermaLink="true">https://flag92.com/en/blog/dify-vs-fastgpt/</guid><description>The two most popular open-source LLM app platforms in China overlap heavily. This compares them across 8 dimensions and gives recommendations.</description><pubDate>Wed, 22 Apr 2026 00:00:00 GMT</pubDate><category>comparison</category><category>Dify</category><category>FastGPT</category></item><item><title>KB preparation in practice — from Notion / Lark / Confluence to RAG</title><link>https://flag92.com/en/blog/knowledge-base-preparation/</link><guid isPermaLink="true">https://flag92.com/en/blog/knowledge-base-preparation/</guid><description>A bad KB ruins any LLM budget. End-to-end repeatable pipeline — export, clean, chunk, augment, evaluate — with Python.</description><pubDate>Wed, 15 Apr 2026 00:00:00 GMT</pubDate><category>practice</category><category>RAG</category><category>knowledge base</category></item><item><title>How to evaluate your RAG quality — 5 metrics, 3 toolkits</title><link>https://flag92.com/en/blog/rag-quality-evaluation/</link><guid isPermaLink="true">https://flag92.com/en/blog/rag-quality-evaluation/</guid><description>&quot;Feels better than last week&quot; is not measurement. Five quantitative RAG metrics and three open-source evaluators.</description><pubDate>Wed, 08 Apr 2026 00:00:00 GMT</pubDate><category>benchmark</category><category>RAG</category><category>quality</category></item><item><title>Production-grade Cloudflare Tunnel tuning for AI support</title><link>https://flag92.com/en/blog/cloudflare-tunnel-tuning/</link><guid isPermaLink="true">https://flag92.com/en/blog/cloudflare-tunnel-tuning/</guid><description>One-command Cloudflare Tunnel for Chatwoot + Dify hides production gotchas — WebSocket drops, timeouts, monitoring blind spots. A checklist.</description><pubDate>Wed, 01 Apr 2026 00:00:00 GMT</pubDate><category>deployment</category><category>Cloudflare</category><category>ops</category></item><item><title>2026 Chinese embedding benchmark — bge-m3, Conan, m3e, bce, OpenAI</title><link>https://flag92.com/en/blog/chinese-embedding-benchmark/</link><guid isPermaLink="true">https://flag92.com/en/blog/chinese-embedding-benchmark/</guid><description>Same Chinese KB, same real support questions — five embedding models compared on retrieval accuracy, speed and cost.</description><pubDate>Tue, 12 May 2026 00:00:00 GMT</pubDate><category>benchmark</category><category>embedding</category><category>RAG</category></item><item><title>Chatwoot deep review — 6 months in production, the unvarnished take</title><link>https://flag92.com/en/blog/chatwoot-deep-review/</link><guid isPermaLink="true">https://flag92.com/en/blog/chatwoot-deep-review/</guid><description>We ran Chatwoot 4.x in production for 6 months — 40k+ tickets, 12 agents, full debrief on Inbox, Captain, API and gotchas.</description><pubDate>Tue, 10 Mar 2026 00:00:00 GMT</pubDate><category>deep-dive</category><category>Chatwoot</category></item><item><title>Dify workflow patterns — 10 we actually use in production</title><link>https://flag92.com/en/blog/dify-workflow-patterns/</link><guid isPermaLink="true">https://flag92.com/en/blog/dify-workflow-patterns/</guid><description>Ten Dify workflow patterns that survive production — KB fallback, context compression, concurrent tool calls, retries and more.</description><pubDate>Tue, 03 Mar 2026 00:00:00 GMT</pubDate><category>deep-dive</category><category>Dify</category><category>workflow</category></item><item><title>Why self-hosting an AI support stack pays off in 2026</title><link>https://flag92.com/en/blog/why-self-host-ai-support/</link><guid isPermaLink="true">https://flag92.com/en/blog/why-self-host-ai-support/</guid><description>SaaS support keeps getting pricier while open-source matures. A framework across cost, compliance and control.</description><pubDate>Sun, 10 May 2026 00:00:00 GMT</pubDate><category>thinking</category><category>evaluation</category></item><item><title>Migrating from Zendesk to Chatwoot — translating tickets and triggers</title><link>https://flag92.com/en/blog/migrate-zendesk-to-chatwoot/</link><guid isPermaLink="true">https://flag92.com/en/blog/migrate-zendesk-to-chatwoot/</guid><description>Zendesk&apos;s ticketing is heavier than Intercom&apos;s; the hardest part of migrating to Chatwoot is rewriting Triggers / Automations / Macros.</description><pubDate>Sun, 25 Jan 2026 00:00:00 GMT</pubDate><category>migration</category><category>Chatwoot</category><category>Zendesk</category></item><item><title>Migrating from OpenAI GPTs to Dify — bring your Custom GPT home</title><link>https://flag92.com/en/blog/migrate-gpts-to-dify/</link><guid isPermaLink="true">https://flag92.com/en/blog/migrate-gpts-to-dify/</guid><description>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.</description><pubDate>Sun, 18 Jan 2026 00:00:00 GMT</pubDate><category>migration</category><category>Dify</category><category>OpenAI</category></item><item><title>Migrating from Intercom to Chatwoot — the complete path</title><link>https://flag92.com/en/blog/migrate-intercom-to-chatwoot/</link><guid isPermaLink="true">https://flag92.com/en/blog/migrate-intercom-to-chatwoot/</guid><description>Data export, field mapping, parallel running, cutover — a battle-tested playbook with scripts and traps.</description><pubDate>Sun, 01 Feb 2026 00:00:00 GMT</pubDate><category>migration</category><category>Chatwoot</category><category>Intercom</category></item><item><title>Case · A cross-border e-commerce running 80k daily orders — 6 months on Chatwoot + Dify</title><link>https://flag92.com/en/blog/case-study-cross-border-ecommerce/</link><guid isPermaLink="true">https://flag92.com/en/blog/case-study-cross-border-ecommerce/</guid><description>A Shopify + WhatsApp e-commerce shop migrated from Intercom to Chatwoot + Dify. Six months of real numbers — cost, deflection, satisfaction.</description><pubDate>Fri, 15 May 2026 00:00:00 GMT</pubDate><category>case study</category><category>e-commerce</category><category>practice</category></item><item><title>Case · A B2B SaaS migrated from Intercom to Chatwoot in 6 weeks</title><link>https://flag92.com/en/blog/case-study-saas-migration/</link><guid isPermaLink="true">https://flag92.com/en/blog/case-study-saas-migration/</guid><description>A 2,000-customer, $2M ARR B2B SaaS completed Intercom → Chatwoot migration in 6 weeks. Full timeline, headcount, scars.</description><pubDate>Fri, 08 May 2026 00:00:00 GMT</pubDate><category>case study</category><category>migration</category><category>SaaS</category><category>Chatwoot</category></item><item><title>2026 Chinese-support LLM bake-off — Qwen, DeepSeek, GLM, Doubao, ERNIE</title><link>https://flag92.com/en/blog/chinese-llm-bakeoff/</link><guid isPermaLink="true">https://flag92.com/en/blog/chinese-llm-bakeoff/</guid><description>Same support prompt and knowledge base — which of the five China-trained LLMs ships the best AI support? 200 real questions decide.</description><pubDate>Fri, 08 May 2026 00:00:00 GMT</pubDate><category>benchmark</category><category>LLM</category><category>model selection</category></item><item><title>Case · A securities firm shipped a fully local AI support — the compliance path</title><link>https://flag92.com/en/blog/case-study-financial-onprem/</link><guid isPermaLink="true">https://flag92.com/en/blog/case-study-financial-onprem/</guid><description>No data egress, regulator approval, internal + external audit, DR plan. The real compliance journey for a mid-sized securities brokerage&apos;s local AI support.</description><pubDate>Fri, 01 May 2026 00:00:00 GMT</pubDate><category>case study</category><category>finance</category><category>compliance</category><category>on-prem</category></item><item><title>Case · An online-edu promo day with 50× traffic — engineering notes</title><link>https://flag92.com/en/blog/case-study-education-promo/</link><guid isPermaLink="true">https://flag92.com/en/blog/case-study-education-promo/</guid><description>A coding-education company handled 120k conversations in 24 hours during 618 sale, AI deflection 81%. How engineering pre-tested, rate-limited, degraded gracefully.</description><pubDate>Fri, 24 Apr 2026 00:00:00 GMT</pubDate><category>case study</category><category>education</category><category>promo day</category><category>load test</category></item><item><title>Case · How a healthcare provider built RAG with zero PHI leakage</title><link>https://flag92.com/en/blog/case-study-healthcare-rag/</link><guid isPermaLink="true">https://flag92.com/en/blog/case-study-healthcare-rag/</guid><description>An internet hospital launched AI health consultations. The hardest part — PHI must never reach the LLM. Their 3-tier KB + Pipeline redaction + audit architecture.</description><pubDate>Fri, 17 Apr 2026 00:00:00 GMT</pubDate><category>case study</category><category>healthcare</category><category>compliance</category><category>RAG</category></item><item><title>Case · An open-source project handled a 300k Discord community with Chatwoot + Dify</title><link>https://flag92.com/en/blog/case-study-discord-community/</link><guid isPermaLink="true">https://flag92.com/en/blog/case-study-discord-community/</guid><description>A 18k-star OSS project went from drowning in Discord to clean AI + volunteer mod tiers handling 300k community asks.</description><pubDate>Fri, 10 Apr 2026 00:00:00 GMT</pubDate><category>case study</category><category>open source</category><category>Discord</category><category>community</category></item></channel></rss>