# Edor.ai

> Edor.ai is an AI school-based learning and teaching platform for Hong Kong primary and secondary schools. It combines AI lesson planning, AI-powered assessment and marking, a scaffolded (Socratic) AI tutor, subject resource libraries with versioning, a knowledge asset base with RAG, a student self-directed learning desk (safe chat, flashcards with SM-2 spaced repetition, flipped learning, games), science inquiry tools, PIE (Plan–Implement–Evaluate) school analytics and SDAS (Student Data Analysis System) — in one secure platform where the school owns its data. Interface: Traditional Chinese (Hong Kong) with English toggle.

Key facts:
- Licensing: per-school annual subscription; unlimited accounts for 800 students and 100 staff; AI usage included, no token fees; includes updates and model upgrades.
- Grant alignment: designed for the Hong Kong EDB 『智』啟學教撥款計劃 (AI Education Grant), launched 16 Dec 2025, funded by the Quality Education Fund, up to HK$500,000 per school, to be used by 31 Aug 2028. Includes teacher training, demo lessons, sharing sessions, on-site consultancy and PIE reporting evidence.
- Safety: four-layer AI guard (input detection, school system prompt, output review, audit + quotas); student mode; human moderation review; the tutor never gives direct answers (hint ladder with teacher-set ceiling).
- Data ownership: the data ALWAYS belongs to the school, whether deployed on-premises (school server / private cloud) or hosted in Edor's cloud. Schools can download everything (JSON / CSV / original files) for free at any time and switch vendors freely — no lock-in. PII fields encrypted (AES-256); TLS; retention policy; student data never used for model training; PDPO-aligned.
- Deployment: cloud hosting at edor.ai or on-premises; schools may connect their own domain (e.g. ai.school.edu.hk) via CNAME with TLS certificates issued and renewed by Edor.
- AI providers: switchable — OpenAI (default), Azure OpenAI, Anthropic, Poe, local Ollama. Per-provider support status is documented at https://edor.ai/en/models.
- Roles: admin, principal, panel head, class teacher, subject teacher, student; subject/class scoping; audit logs.
- Mobile: Progressive Web App (add to home screen on iOS/Android), camera upload, voice input.
- Contact: info@edor.ai · https://edor.ai/contact

## Pages (zh-HK)
- [首頁](https://edor.ai/): platform overview
- [功能](https://edor.ai/features): nine modules in detail
- [『智』啟學教撥款](https://edor.ai/ai-funding): grant facts, timeline, deliverables mapping, FAQ
- [安全與合規](https://edor.ai/security): security architecture, tender requirement mapping A1–A5, data ownership & portability, privacy summary
- [方案](https://edor.ai/pricing): Primary, Secondary (with SDAS), School Sponsoring Body, on-premises option
- [常見問題](https://edor.ai/faq)
- [教育 AI 專欄](https://edor.ai/blog): articles on AI in education for Hong Kong schools
- [AI 模型資料庫](https://edor.ai/models): specs, pricing and platform support status for major LLM providers
- [學習中心](https://edor.ai/learn): free AI literacy courses for teachers (LLM fundamentals, prompt engineering)
- [聯絡我們](https://edor.ai/contact)

## Pages (English)
- [Home](https://edor.ai/en)
- [Features](https://edor.ai/en/features)
- [AI Education Grant](https://edor.ai/en/ai-funding)
- [Security](https://edor.ai/en/security)
- [Pricing](https://edor.ai/en/pricing)
- [FAQ](https://edor.ai/en/faq)
- [Blog](https://edor.ai/en/blog)
- [AI Model Database](https://edor.ai/en/models)
- [Learning Centre](https://edor.ai/en/learn)
- [Contact](https://edor.ai/en/contact)

## Blog articles (zh-HK)
- [2026 年香港教師最常問的九個 AI 工具:各自真正的用途、價格與學生能否使用](https://edor.ai/blog/ai-tools-for-teachers-2026-overview): Gemini、Gamma、Edpuzzle、Khanmigo、Perplexity、OpenAI Codex、Claude Code、Cursor、Base44——九個工具,九種完全不同的用途。本文以 2026 年 9 月的官方資料,整理每個工具真正為誰而設、免費版能做甚麼、學生在條款上能否使用,並提供一份按任務分類的選擇清單,以及通用工具無論如何都不會替學校處理的四件事。
- [2026 全模型橫評:15 家供應商、一套學校用的選型框架](https://edor.ai/blog/all-llm-models-compare-2026): 由 OpenAI、Anthropic、Google、xAI、Meta、Mistral,到 DeepSeek、Qwen、Kimi、GLM、MiniMax、豆包、混元、文心,再加上編碼專用的 Cursor Composer。本文以同一套學校用量估算,把 15 家供應商的價格、上下文、開放權重與實際可用性放在同一張表上,並提供一套四問選型框架。
- [Base44 教師完整教學(2026 年 9 月):用一句話做一個學校網頁應用,以及它的 credits 陷阱](https://edor.ai/blog/base44-for-educators-guide): Base44 是 Wix 旗下的 vibe coding 應用程式生成器,教師不需要寫程式,描述需求就能生成有資料庫、登入與網頁寄存的應用。本文整理 2026 年 9 月官方方案(免費、Starter 每月 16 美元、Builder 40 美元、Pro 80 美元、Elite 160 美元,均為年繳價)、message credits 與 integration credits 的分別、13 歲年齡門檻,以及一個小五閱讀紀錄追蹤應用的完整製作流程。
- [中國大模型橫評:DeepSeek、Qwen、Kimi、GLM 等八家,香港學校可以怎樣用?](https://edor.ai/blog/china-llm-models-compare-deepseek-kimi-qwen-glm): DeepSeek 的價格是 GPT-6 Astra 的六十七分之一,中文能力亦不輸人。那為甚麼校本平台不直接接上去?本文橫評 DeepSeek、Qwen、Kimi、GLM、MiniMax、豆包、混元與文心八家,說明開放權重與雲端 API 的關鍵分別,以及《個人資料(私隱)條例》下香港學校真正可行的三條路線。
- [Claude Code 教師完整教學(2026 年 9 月):方案價格、18 歲限制與自動測試學生程式](https://edor.ai/blog/claude-code-for-educators-guide): Claude Code 是 Anthropic 的代理式編程工具,包含在所有付費 Claude 方案內,免費方案不提供。本文整理 2026 年 9 月官方價格(Pro 每月 20 美元、Max 由 100 美元起、Team 標準席位每人 20 至 25 美元、Enterprise 每席位 20 美元加用量)、五小時滾動用量視窗的計算方式、Claude 18 歲政策對學生的影響,以及一個為中六資訊科技科建立自動測試腳本的完整流程。
- [Cursor 教師完整教學(2026 年 9 月):價格、學生帳戶限制與資訊科技科實戰](https://edor.ai/blog/cursor-for-educators-guide): Cursor 是 Anysphere 的 AI 程式編輯器,2026 年 9 月方案為 Hobby 免費、Pro 每月 20 美元、Pro+ 60 美元、Ultra 200 美元、Teams 每人 40 美元、Teams Premium 每人 120 美元。本文拆解自家模型 Composer 2.5 的定價、Teams 的 Cursor Token Rate 附加費、18 歲年齡門檻對學生帳戶的影響,以及一個中三資訊科技科批改 Python 作業的完整操作流程。
- [Edpuzzle 教師完整教學:用互動影片翻轉初中科學課,以及 20 條活動上限的真相](https://edor.ai/blog/edpuzzle-for-educators-guide): Edpuzzle 對教師免費,但免費 Basic 方案只能儲存 20 個活動。本文以中二科學科翻轉課堂示範完整流程,包括自動出題、Autograde、Supported Assignments 與防抄襲設定,並拆解 Pro Teacher 與 Pro School 的真實成本、家長同意要求,以及它沒有繁體中文介面這件事。
- [由聊天機械人到 AI 代理:工具呼叫、多步驟自動化與 MCP,學校應該自動化甚麼?](https://edor.ai/blog/from-chatbot-to-ai-agent-tool-calling): 聊天機械人只會說話,AI 代理會動手:查資料、讀文件、呼叫系統、連續執行多個步驟。本文用教師看得懂的方式解釋工具呼叫與 MCP 標準,對照各家的工具收費,並提出一份「應該自動化 / 不應該自動化」清單——關鍵不是 AI 能做甚麼,而是哪一步必須由人按下確認。
- [Gamma 教師完整教學:15 分鐘做一份家長晚會簡報,以及那個 16 歲年齡門檻](https://edor.ai/blog/gamma-app-for-educators-guide): Gamma 用 credits 而不是月費計算 AI 用量,免費帳戶得到 400 個一次性 credits 且永不補充。本文以香港家長晚會簡報示範完整流程,拆解 credits 怎樣燒、100 位教師按人頭訂閱的真實成本,並指出官方條款要求使用者至少 16 歲——這一句決定了它能不能給學生用。
- [Google Gemini 教師完整教學:Workspace for Education、Gemini in Classroom 與真實收費拆解](https://edor.ai/blog/gemini-for-educators-guide): Gemini for Education 在所有 Google Workspace for Education 版本中免費,但 Docs、Gmail 內的 Gemini 要另購 Google AI Pro for Education,每位使用者每月 20 美元且限 18 歲以上。本文以香港教師的分層工作紙情境示範完整流程,並拆解 100 位教職員的實際成本、學生年齡限制與繁體中文支援。
- [Khanmigo 香港教師完整教學:教師工具免費可用,但學生導師香港買不到](https://edor.ai/blog/khanmigo-for-educators-guide): Khanmigo for Teachers 已開放給 180 多個國家及地區的教師免費使用,香港教師符合資格;但付費的 Khanmigo 學生與家長訂閱(每月 4 美元、每年 44 美元)官方明確只在美國提供。本文示範用教師工具建立數學引導式輔導流程、檢視學生近七天表現,並拆解香港學校能與不能做的事。
- [四大美國模型供應商橫評:OpenAI、Anthropic、xAI、Google 哪一家適合學校?](https://edor.ai/blog/llm-providers-compare-openai-anthropic-xai-google): 2026 年 9 月,四家美國前沿供應商的旗艦價格由每 100 萬 token 0.75 美元到 10 美元,相差 13 倍。本文以香港學校最關心的六條軸線(價格、上下文、中文處理、內容安全、平台可用性、成本陷阱)做橫評,並解釋為甚麼「最強的模型」通常不是學校應該用的模型。
- [安全、幻覺與風險控制:提示注入、幻覺後果,以及校本平台必須有的護欄](https://edor.ai/blog/llm-safety-hallucination-prompt-injection-risk): AI 在通告草稿裏引用了一個格式完全正確、但根本不存在的課程指引編號。這類錯誤不是偶發,而是語言模型的結構性特徵。本文用 OWASP 十大風險與 NIST 風險管理框架為骨架,逐項對照香港學校的真實場景,並列出一個校本 AI 平台必須具備的護欄與稽核安排。
- [開源權重還是閉源 API?自架與租用的成本真帳,以及私隱在哪一刻改變答案](https://edor.ai/blog/open-source-vs-closed-llm-cost-privacy): 很多學校以為「自己買機跑開源模型」一定慳錢。把真實用量代進 2026 年 9 月的官方價格後,答案往往相反:雲端 API 的年費低到硬件根本追不上。本文算清兩邊的真帳,列出對學校現實可行的開源型號與硬件門檻,並說明私隱與資料政策在哪一刻會令天秤倒向自架。
- [OpenAI Codex 教師完整教學(2026 年 9 月):方案、額度制收費與課堂小工具實作](https://edor.ai/blog/openai-codex-for-educators-guide): Codex 是 OpenAI 的代理式編程產品,包含終端機 CLI、IDE 擴充、雲端任務與 iOS 版本,已包含在 ChatGPT Free、Go、Plus、Pro、Business、Edu 與 Enterprise 方案內。本文整理 2026 年 9 月的官方價格、credits 額度費率、GPT-5.6 與 GPT-6 Astra 的用量差異、美國 K-12 教師免費計劃為何香港用不到,以及一個教師自製課堂小工具的完整 CLI 操作流程。
- [Perplexity AI 教師完整教學:公民與社會發展科專題研習,以及教學生核對引用](https://edor.ai/blog/perplexity-ai-for-educators-guide): Perplexity 每個答案都附來源,是教「查證」最好用的工具。Comet 瀏覽器免費,Pro 每月 20 美元、年繳每月 17 美元,Enterprise Pro 每席位每月 40 美元。本文示範一個香港公社科專題研習流程、一張學生用的引用核對表,並指出它 13 歲的年齡條款與沒有教師監控這兩個限制。
- [推理革命:由思維鏈到自適應思考,對評卷、數學與科學教學改變了甚麼?](https://edor.ai/blog/reasoning-revolution-chain-of-thought-llm): 2024 年的模型是「脫口而出」,2026 年的模型會先想清楚才答,而且自己決定想多久。本文用教師看得懂的方式解釋思維鏈、推理力度與自適應思考的演變,對照四家供應商的調節掣設計,並說明它在評卷、數學與科學教學上真正改變了甚麼——以及仍然改變不了甚麼。
- [香港學校 AI 教育入門:2026 校本 AI 平台該具備的 10 個條件](https://edor.ai/blog/ai-education-platform-10-requirements): 為校長與 IT 主任整理選擇校本 AI 學與教平台的 10 個必備條件——由課程對接、學生安全、教師監控、數據擁有權到撥款合規,附可直接用於招標的檢視表。
- [『智』啟學教撥款 HK$50 萬怎樣用最有價值?承諾項目與時間表全解](https://edor.ai/blog/ai-education-grant-500k-how-to-use): 一文看懂教育局『智』啟學教撥款計劃:資助金額、使用期限、四項承諾(3 科 × 2 級、6 個教學示例、3 節示範課、3 次分享會)、中期與最終報告時間,以及一份可直接套用的預算分配建議。
- [學校選購 AI 平台的 12 個評估問題:數據擁有權、隨時全量下載、可自由轉換供應商](https://edor.ai/blog/12-questions-before-buying-school-ai-platform): 給校長、IT 主任與採購小組的 12 條問題:由數據擁有權、匯出格式、無鎖定期、token 收費、多模型、自有網域到本地/雲端部署,每題附「合格答案」與「警號」。
- [本地部署 vs 雲端部署:學校該怎樣選?兩者數據都屬於學校,都可隨時整套下載](https://edor.ai/blog/on-premises-vs-cloud-deployment-schools): 比較 AI 學與教平台的本地部署(校內伺服器 / 私有雲)與雲端代管:成本、維護、更新速度、資料所在地、合規與可攜性。結論先講:無論哪一種,數據都屬於學校、隨時可整套下載,並可用學校自有網域接入。
- [AI 備課實戰:教師如何用 AI 在 30 分鐘完成一份教案與分層工作紙](https://edor.ai/blog/ai-lesson-planning-30-minutes): 以小四中文「說明文」與中二數學「一元一次方程」為例,示範教師如何以校本課程單元為起點,用 AI 在 30 分鐘內完成教案、簡報大綱、三層工作紙與課後反思,並附可直接套用的提示結構。
- [鷲架式(Scaffolding)智能導師:為什麼「不直接給答案」的 AI 才適合學生](https://edor.ai/blog/scaffolding-ai-tutor-no-direct-answers): 解釋鷲架式智能導師的 4 級提示階梯、蘇格拉底式提問與教師「提示上限」設計,說明為何直接給答案的 AI 會削弱學習,並附小學數學與中學英文的對話示例。
- [AI 出題評卷與即時回饋:數學、中文、英文寫作的實踐與限制](https://edor.ai/blog/ai-marking-instant-feedback): 拆解 AI 評卷在三個科目的做法——數學步驟批改(含手寫拍照)、中文作文按評分準則回饋、英文寫作語法與結構建議——以及教師必須保留的三個把關點,附評分準則範本。
- [用 AI 做差異化教學:同一課題如何自動生成三層難度與 SEN 調適版本](https://edor.ai/blog/differentiated-instruction-with-ai): 差異化教學過去最大的障礙是「時間」。本文示範如何用 AI 在同一學習目標下生成基礎、核心、挑戰三層任務,以及讀寫障礙、ADHD、非華語學生的調適版本,並提供教師分配與追蹤的實務流程。
- [間隔重複 + AI 溫習卡:讓學生記得更久的科學方法](https://edor.ai/blog/spaced-repetition-ai-flashcards): 解釋間隔重複(SM-2 演算法)為何比考前狂溫有效,示範 AI 如何由課文自動生成溫習卡、按學生表現安排複習時間,以及教師如何用溫習數據找出全班最弱的概念。
- [翻轉課室 2.0:AI 自動生成影片摘要、字幕與課前檢查問題](https://edor.ai/blog/flipped-classroom-2-ai-video-summary): 翻轉課室失敗的常見原因是學生「看了但沒吸收」。本文示範如何用 AI 把任何教學影片轉成字幕、分段摘要、3–5 條檢查問題與迷思提示,並在課前收集學生回應,讓課堂時間真正用於討論與應用。
- [AI 口語練習:小學英文與普通話說話課如何用語音辨識與朗讀每天練](https://edor.ai/blog/ai-oral-practice-primary-english-putonghua): 說話能力需要「每天開口」,但一位教師無法逐一聆聽 30 位學生。本文示範 AI 口語練習的流程——情境對話、語音辨識(STT)、朗讀示範(TTS)、發音與流暢度回饋——以及教師如何以練習紀錄設計下一課。
- [科學探究(STEAM)課用 AI 設計實驗與探究活動:由問題到紀錄表](https://edor.ai/blog/science-inquiry-steam-ai-experiment-design): 配合小學科學科新課程,示範教師如何用 AI 在 20 分鐘內把一個課題變成完整探究活動——探究問題、假設、變項、步驟、安全提示、紀錄表與延伸——並提供三個實例與一份「探究品質檢查表」。
- [學生數據分析(SDAS):從成績表到跨年追蹤,校長與科主任要問的 8 個問題](https://edor.ai/blog/sdas-student-data-analysis-8-questions): 成績表只告訴你「誰考了多少分」;學生數據分析系統(SDAS)回答的是「為什麼、趨勢如何、下一步做什麼」。本文以 8 個問題示範標準分數、百分位、進步率、跨年追蹤與教師分析的用法,以及 AI 自然語言查詢如何讓非數據背景的教師也能用。
- [AI 如何減輕教師行政負擔:通告、會議紀錄、家長信與 eAdmin 實例](https://edor.ai/blog/reduce-teacher-admin-workload-ai-eadmin): 教師每週花在行政文書的時間可達 8–10 小時。本文以六個實例示範 AI 行政代理(eAdmin)如何草擬 eNotice、eCircular、家長信、會議紀錄、活動計劃書與 iPortfolio 評語,並說明「引用校本知識庫」與「符合學校語氣」為何是關鍵。
- [PIE(策劃—推行—檢討)循環遇上 AI 儀表板:校務決策如何有數據依據](https://edor.ai/blog/pie-cycle-ai-dashboard-school-planning): 學校每年都做 PIE,但「檢討」往往靠印象。本文示範如何把 PIE 的 KPI 直接連結平台數據——AI 使用量、學習目標掌握度、教師培訓紀錄、SDAS 成績——讓策劃有基線、推行有追蹤、檢討有證據,並自動產出撥款中期與最終報告。
- [學生用 AI 的安全網:四層守門、個人資料保護與教師監察怎樣設計](https://edor.ai/blog/student-ai-safety-guardrails-pii-monitoring): 讓小學生使用 AI,學校最擔心三件事:不當內容、個人資料外洩、無人監察。本文拆解一套可落地的安全架構——輸入守門、系統提示、輸出審核、審計配額——以及 PII 加密、不訓練條款、人工複核與 PDPO 對照的具體做法。
- [校本知識庫(RAG):讓 AI 引用學校自己的教材與政策,並標明來源](https://edor.ai/blog/school-knowledge-base-rag-ai): 通用 AI 不知道你學校的校規、課程進度與通告格式。校本知識庫以 RAG(檢索增強生成)讓 AI 在回答前先檢索學校文件,並在答案中標明來源。本文說明原理、應放什麼文件、權限與匯出,以及教師與行政的六個應用場景。

## Blog articles (English)
- [Nine AI Tools Hong Kong Teachers Keep Asking About in 2026: What Each Is Really For, What It Costs, and Whether Students May Use It](https://edor.ai/en/blog/ai-tools-for-teachers-2026-overview): Gemini, Gamma, Edpuzzle, Khanmigo, Perplexity, OpenAI Codex, Claude Code, Cursor and Base44 — nine tools with nine entirely different purposes. Using official September 2026 sources, this hub sets out who each one is really built for, what the free tier allows, whether students are permitted under the terms, a which-tool-for-which-job decision list, and the four things no general-purpose tool will ever do for a school.
- [The 2026 All-Model Roundup: 15 Providers and One Selection Framework for Schools](https://edor.ai/en/blog/all-llm-models-compare-2026): From OpenAI, Anthropic, Google, xAI, Meta and Mistral to DeepSeek, Qwen, Kimi, GLM, MiniMax, Doubao, Hunyuan and ERNIE, plus the coding-only Cursor Composer. One school usage model, applied to every vendor's published pricing, puts all 15 on a single table — followed by a four-question framework for choosing between them.
- [Base44 for Teachers: The Complete Guide (September 2026) — Build a School Web App by Describing It, and the Credit Trap](https://edor.ai/en/blog/base44-for-educators-guide): Base44 is Wix's vibe-coding app builder, letting teachers create an app with a database, login and hosting without writing code. This guide covers the September 2026 official plans (Free, Starter US$16/month, Builder US$40, Pro US$80, Elite US$160, all annual), the difference between message credits and integration credits, the age-13 minimum, and a full walkthrough building a Primary 5 reading-log tracker.
- [China's LLMs Compared: DeepSeek, Qwen, Kimi, GLM and Five More — What Can a Hong Kong School Actually Use?](https://edor.ai/en/blog/china-llm-models-compare-deepseek-kimi-qwen-glm): DeepSeek costs one sixty-seventh of GPT-6 Astra per input token and is no weaker in Chinese. So why does a school platform not simply plug into it? This roundup compares DeepSeek, Qwen, Kimi, GLM, MiniMax, Doubao, Hunyuan and ERNIE, explains the crucial difference between open weights and a cloud API, and sets out the three routes that are genuinely workable under the Personal Data (Privacy) Ordinance.
- [Claude Code for Teachers: The Complete Guide (September 2026) — Pricing, the 18+ Policy and Auto-testing Student Code](https://edor.ai/en/blog/claude-code-for-educators-guide): Claude Code is Anthropic's agentic coding tool, included with every paid Claude plan and unavailable on Free. This guide covers the September 2026 official prices (Pro US$20/month, Max from US$100, Team standard seats US$20–25 per user, Enterprise US$20 per seat plus usage), how the rolling five-hour usage window works, what Claude's 18-and-over policy means for pupils, and a full walkthrough building an auto-test harness for a Form 6 ICT class.
- [Cursor for Teachers: The Complete Guide (September 2026) — Pricing, the 18+ Rule, and Marking Student Code](https://edor.ai/en/blog/cursor-for-educators-guide): Cursor is Anysphere's AI code editor. As of September 2026 the plans are Hobby free, Pro US$20/month, Pro+ US$60, Ultra US$200, Teams US$40/user and Teams Premium US$120/user. This guide covers Composer 2.5 pricing, the Cursor Token Rate surcharge on Teams, why the 18-and-over terms rule out student accounts, and a full walkthrough of marking a Form 3 ICT class's Python homework.
- [Edpuzzle for Teachers: Flipping a Junior Science Lesson, and the Truth About the 20-Activity Cap](https://edor.ai/en/blog/edpuzzle-for-educators-guide): Edpuzzle is free for teachers, but the free Basic plan stores only 20 activities. This guide flips a Secondary 2 science lesson step by step, covering auto-generated questions, Autograde, Supported Assignments and anti-cheating settings, then breaks down what Pro Teacher and Pro School really cost, the parental consent duty, and the fact that there is no Traditional Chinese interface.
- [From Chatbot to AI Agent: Tool Calling, Multi-Step Automation, MCP, and What a School Should Not Automate](https://edor.ai/en/blog/from-chatbot-to-ai-agent-tool-calling): A chatbot only talks. An agent acts — it looks things up, reads documents, calls systems and runs several steps in sequence. This article explains tool calling and the MCP standard in terms a teacher can use, compares published tool-call charges, and offers a should-automate and should-not-automate list. The question is not what AI can do, but which step a person must confirm.
- [Gamma for Teachers: A Parents' Evening Deck in 15 Minutes, and the Age-16 Clause](https://edor.ai/en/blog/gamma-app-for-educators-guide): Gamma meters AI usage in credits rather than a flat monthly allowance, and a free account gets 400 one-off credits that never refresh. This guide builds a Hong Kong parents' evening deck step by step, breaks down how credits actually burn, works out the per-seat cost across 100 staff, and flags that Gamma's terms require users to be at least 16.
- [Google Gemini for Teachers: A Full Tutorial, Workspace for Education, Gemini in Classroom and the Real Bill](https://edor.ai/en/blog/gemini-for-educators-guide): Gemini for Education is included at no cost in every Google Workspace for Education edition, but the Gemini sidebar inside Docs and Gmail needs the Google AI Pro for Education add-on at US$20 per user per month, and that add-on is restricted to users aged 18 and over. This guide walks a Hong Kong teacher through building a differentiated Chinese worksheet, then breaks down what 100 staff licences really cost.
- [Khanmigo for Hong Kong Teachers: The Teacher Tools Are Free Here, the Student Tutor Cannot Be Bought](https://edor.ai/en/blog/khanmigo-for-educators-guide): Khanmigo for Teachers is now free to educators in more than 180 countries and territories and Hong Kong qualifies, but the paid Khanmigo learner and parent subscriptions at US$4 a month or US$44 a year are officially available only in the United States. This guide builds a Secondary 1 maths hint ladder with the teacher tools, reviews what pupils have been working on, and states plainly what a Hong Kong school can and cannot do.
- [OpenAI vs Anthropic vs xAI vs Google: Which US Provider Actually Suits a School?](https://edor.ai/en/blog/llm-providers-compare-openai-anthropic-xai-google): In September 2026 the four leading US flagships range from $0.75 to $10 per 1M input tokens — a 13x spread. This head-to-head compares them on the six axes a school actually cares about (price, context, Chinese-language handling, content safety, platform availability and cost cliffs) and explains why the strongest model is usually not the one a school should be running.
- [Safety, Hallucination and Risk Control: Prompt Injection, Real Consequences, and the Guardrails a School Platform Needs](https://edor.ai/en/blog/llm-safety-hallucination-prompt-injection-risk): An AI drafted a parent notice citing a curriculum guideline number that was perfectly formatted and did not exist. That is not a glitch; it is a structural property of language models. Using the OWASP Top 10 and the NIST AI Risk Management Framework as scaffolding, this article maps each risk to a real Hong Kong school scenario and sets out the guardrails and audit arrangements a school AI platform must have.
- [Open Weights or Closed API? The Real Cost Maths of Self-Hosting, and Where Privacy Flips the Answer](https://edor.ai/en/blog/open-source-vs-closed-llm-cost-privacy): Many schools assume that buying a machine to run an open model must be cheaper. Put real usage through September 2026 published pricing and the answer usually reverses — the annual cloud API bill is too small for hardware to beat. This article works through both sides honestly, lists the open models that are realistic on school hardware, and identifies the point at which data policy tips the balance towards self-hosting.
- [OpenAI Codex for Teachers: The Complete Guide (September 2026) — Plans, Credit Pricing and Building a Classroom Utility](https://edor.ai/en/blog/openai-codex-for-educators-guide): Codex is OpenAI's agentic coding product — a terminal CLI, IDE extension, cloud tasks and an iOS app — included with ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise. This guide covers the September 2026 official prices, the credit rate card, the difference between GPT-5.6 and GPT-6 Astra usage, why the free US K-12 teacher plan does not reach Hong Kong, and a full CLI walkthrough for building a classroom utility.
- [Perplexity AI for Teachers: A Sourced Project Research Flow, and Teaching Pupils to Check Citations](https://edor.ai/en/blog/perplexity-ai-for-educators-guide): Every Perplexity answer carries its sources, which makes it the best available tool for teaching verification. The Comet browser is free, Pro is US$20 a month or US$17 a month billed annually, and Enterprise Pro is US$40 per seat per month. This guide runs a Hong Kong Citizenship and Social Development project, supplies a citation-checking sheet for pupils, and flags the age-13 clause and the absence of teacher monitoring.
- [The Reasoning Revolution: From Chain-of-Thought to Adaptive Thinking, and What It Changes for Marking, Maths and Science](https://edor.ai/en/blog/reasoning-revolution-chain-of-thought-llm): Models in 2024 blurted out an answer. Models in 2026 think first, and decide for themselves how long to think. This article explains chain-of-thought, reasoning effort and adaptive thinking in terms a teacher can use, compares the four vendors' effort dials, and sets out what has genuinely changed for marking, maths and science teaching — and what has not.
- [AI in Hong Kong Schools: 10 Requirements a School-Based AI Platform Must Meet in 2026](https://edor.ai/en/blog/ai-education-platform-10-requirements): A checklist for principals and IT coordinators choosing a school-based AI learning platform — curriculum alignment, student safety, teacher oversight, data ownership and grant compliance, ready to paste into a tender.
- [How to Get the Most Value from the HK$500,000 AI Education Grant: Commitments, Timeline and Budget](https://edor.ai/en/blog/ai-education-grant-500k-how-to-use): The EDB AI Education Grant (『智』啟學教) explained — amount, deadline, the four commitments (3 subjects × 2 grades, 6 teaching exemplars, 3 demo lessons, 3 sharing sessions), report dates, and a ready-to-adapt budget split.
- [12 Questions to Ask Before Buying a School AI Platform: Data Ownership, Full Export, Freedom to Switch](https://edor.ai/en/blog/12-questions-before-buying-school-ai-platform): For principals, IT coordinators and procurement teams — 12 questions covering data ownership, export formats, lock-in, token fees, multi-model support, custom domains and on-premises vs cloud, each with a "passing answer" and "red flag".
- [On-Premises vs Cloud Deployment for Schools: How to Choose — the Data Belongs to the School Either Way](https://edor.ai/en/blog/on-premises-vs-cloud-deployment-schools): Comparing on-premises (school server / private cloud) and cloud hosting for an AI learning platform — cost, maintenance, update speed, data location, compliance and portability. Spoiler — with either option the data belongs to the school, can be downloaded in full at any time, and can be reached through the school's own domain.
- [AI Lesson Planning in Practice: A Lesson Plan and Tiered Worksheets in 30 Minutes](https://edor.ai/en/blog/ai-lesson-planning-30-minutes): Using P4 Chinese (expository writing) and S2 Mathematics (linear equations) as examples, see how teachers start from a school curriculum unit and use AI to produce a lesson plan, slide outline, three-tier worksheets and reflection prompts in 30 minutes — with a reusable prompt structure.
- [The Scaffolded AI Tutor: Why an AI That Won't Give Answers Is the Right One for Students](https://edor.ai/en/blog/scaffolding-ai-tutor-no-direct-answers): The 4-level hint ladder, Socratic questioning and teacher-set "hint ceiling" behind a scaffolded AI tutor — why answer-giving AI weakens learning, with dialogue examples from primary maths and secondary English.
- [AI Question Generation, Marking and Instant Feedback: Practice and Limits in Maths, Chinese and English Writing](https://edor.ai/en/blog/ai-marking-instant-feedback): How AI marking works in three subjects — step-by-step maths marking (including handwritten photos), Chinese composition feedback against a rubric, English writing grammar and structure suggestions — plus the three checkpoints teachers must keep, with a rubric template.
- [Differentiated Instruction with AI: Three Difficulty Tiers and SEN Adaptations from One Topic, Automatically](https://edor.ai/en/blog/differentiated-instruction-with-ai): The biggest barrier to differentiation has always been time. This article shows how AI generates foundation, core and challenge tasks under one learning objective, plus adapted versions for dyslexia, ADHD and non-Chinese-speaking students, with a practical flow for assigning and tracking.
- [Spaced Repetition + AI Flashcards: The Science of Helping Students Remember Longer](https://edor.ai/en/blog/spaced-repetition-ai-flashcards): Why spaced repetition (the SM-2 algorithm) beats last-minute cramming, how AI generates flashcards from lesson texts and schedules reviews by each student's performance, and how teachers use review data to find the class's weakest concepts.
- [Flipped Classroom 2.0: AI-Generated Video Summaries, Subtitles and Pre-Lesson Check Questions](https://edor.ai/en/blog/flipped-classroom-2-ai-video-summary): Flipped classrooms usually fail because students "watched but did not absorb". See how AI turns any teaching video into subtitles, segment summaries, 3–5 check questions and misconception alerts, and collects student responses before class so lesson time goes to discussion and application.
- [AI Oral Practice: Daily Speaking Practice for Primary English and Putonghua with Speech Recognition and Read-Aloud](https://edor.ai/en/blog/ai-oral-practice-primary-english-putonghua): Speaking improves only by speaking every day, yet one teacher cannot listen to 30 students individually. This article walks through AI oral practice — scenario dialogues, speech-to-text, read-aloud models, pronunciation and fluency feedback — and how teachers use practice records to plan the next lesson.
- [Designing Science Inquiry (STEAM) Activities with AI: From Question to Record Sheet](https://edor.ai/en/blog/science-inquiry-steam-ai-experiment-design): Aligned with the new primary Science curriculum, this article shows how teachers turn a topic into a complete inquiry activity in 20 minutes with AI — question, hypothesis, variables, procedure, safety notes, record sheet and extension — with three worked examples and an inquiry quality checklist.
- [Student Data Analysis (SDAS): From Mark Sheets to Multi-Year Tracking — 8 Questions Principals and Panel Heads Should Ask](https://edor.ai/en/blog/sdas-student-data-analysis-8-questions): A mark sheet tells you who scored what; a Student Data Analysis System answers why, what the trend is and what to do next. Eight questions show how standard scores, percentiles, progress rates, multi-year tracking and teacher analysis are used — and how AI natural-language queries let every teacher ask.
- [How AI Reduces Teacher Admin Workload: Notices, Minutes, Parent Letters and eAdmin in Practice](https://edor.ai/en/blog/reduce-teacher-admin-workload-ai-eadmin): Teachers can spend 8–10 hours a week on paperwork. Six examples show how AI admin agents (eAdmin) draft eNotice, eCircular, parent letters, meeting minutes, event proposals and iPortfolio comments — and why grounding in the school knowledge base and matching the school's voice are the keys.
- [The PIE Cycle Meets AI Dashboards: Evidence-Based School Decisions](https://edor.ai/en/blog/pie-cycle-ai-dashboard-school-planning): Every school runs Plan–Implement–Evaluate, but "evaluate" often relies on impressions. See how PIE KPIs link directly to platform data — AI usage, objective mastery, teacher training records, SDAS results — so planning has a baseline, implementation has tracking, evaluation has evidence, and grant interim and final reports come out automatically.
- [A Safety Net for Students Using AI: Designing Four-Layer Guardrails, Personal Data Protection and Teacher Monitoring](https://edor.ai/en/blog/student-ai-safety-guardrails-pii-monitoring): When primary pupils use AI, schools worry about three things — inappropriate content, personal data leaks and no oversight. This article lays out a workable architecture — input guard, system prompt, output review, audit & quotas — plus PII encryption, no-training terms, human review and PDPO mapping.
- [The School Knowledge Base (RAG): Letting AI Cite Your Own Materials and Policies](https://edor.ai/en/blog/school-knowledge-base-rag-ai): Generic AI does not know your school rules, pacing or notice format. A school knowledge base uses RAG (retrieval-augmented generation) so the AI searches school documents before answering and cites its sources. This article covers how it works, what to upload, permissions and export, and six use cases for teaching and administration.

## Model reference pages
Independent, source-cited reference pages. Edor.ai has no partnership with these vendors; prices follow the vendors' own published documentation.
- [OpenAI — GPT-6 Astra](https://edor.ai/en/models/openai): released 2026-09-03; Edor.ai support: Native support. Official docs: https://developers.openai.com/api/docs/models/gpt-6-astra
- [Anthropic — Claude Fable 5.1](https://edor.ai/en/models/anthropic-claude): released 2026-09-01; Edor.ai support: Native support. Official docs: https://platform.claude.com/docs/en/models/fable-5-1/overview
- [Google — Gemini 3.8 Flash](https://edor.ai/en/models/google-gemini): released 2026-09-02; Edor.ai support: Via Poe. Official docs: https://ai.google.dev/gemini-api/docs/models
- [xAI — Grok 4.6](https://edor.ai/en/models/xai-grok): released 2026-08-12; Edor.ai support: Via Poe. Official docs: https://docs.x.ai/developers/grok-4-6
- [Meta — Muse Glimmer 30B](https://edor.ai/en/models/meta): released 2026-08-09; Edor.ai support: Self-hosted via Ollama. Official docs: https://huggingface.co/meta-models
- [Mistral AI — Mistral Large 3](https://edor.ai/en/models/mistral): released 2025-12; Edor.ai support: Self-hosted via Ollama. Official docs: https://docs.mistral.ai/
- [Cursor (Anysphere) — Composer 2.5](https://edor.ai/en/models/cursor-composer): released 2026-05-18; Edor.ai support: Not supported. Official docs: https://cursor.com/docs/models/cursor-composer-2-5
- [DeepSeek — DeepSeek-V4.1-Flash](https://edor.ai/en/models/deepseek): released 2026-09-10; Edor.ai support: Self-hosted via Ollama. Official docs: https://api-docs.deepseek.com/
- [Alibaba Qwen — Qwen3.8-Max](https://edor.ai/en/models/alibaba-qwen): released 2026-09-03; Edor.ai support: Self-hosted via Ollama. Official docs: https://huggingface.co/Qwen
- [Moonshot AI (Kimi) — Kimi K3](https://edor.ai/en/models/moonshot-kimi): released 2026-07-16; Edor.ai support: Not supported. Official docs: https://platform.kimi.ai/
- [Zhipu AI / Z.ai (GLM) — GLM-5.3](https://edor.ai/en/models/zhipu-glm): released 2026-08-14; Edor.ai support: Self-hosted via Ollama. Official docs: https://docs.z.ai/
- [MiniMax — MiniMax M3](https://edor.ai/en/models/minimax): released 2026-06; Edor.ai support: Not supported. Official docs: https://huggingface.co/MiniMaxAI
- [ByteDance Doubao — Doubao Seed 2.1 Pro](https://edor.ai/en/models/bytedance-doubao): released 2026-06-23; Edor.ai support: Not supported. Official docs: https://www.volcengine.com/docs/82379
- [Tencent Hunyuan — Hunyuan Hy4 Preview](https://edor.ai/en/models/tencent-hunyuan): released 2026-08-28; Edor.ai support: Self-hosted via Ollama. Official docs: https://huggingface.co/tencent
- [Baidu ERNIE — ERNIE 5.1](https://edor.ai/en/models/baidu-ernie): released 2026-08; Edor.ai support: Not supported. Official docs: https://cloud.baidu.com/doc/qianfan/index.html

## Learning centre (free courses for teachers)
- [LLM Classroom: large language model foundations](https://edor.ai/en/learn/llm): 28 free lessons, about 345 minutes total. From the Transformer architecture, tokenisation, parameters and self-attention through the three training stages, RAG, quantisation, local deployment, AI guardrails and future trends. Everything explained with everyday analogies — no maths or coding background required.
  - [Why teachers need to understand LLMs, not just use them](https://edor.ai/en/learn/llm/why-teachers-should-understand-llm)
  - [The Transformer architecture in plain language](https://edor.ai/en/learn/llm/transformer-architecture-plain-language)
  - [Tokenisation, and why Chinese costs more than English](https://edor.ai/en/learn/llm/tokenization-and-why-chinese-costs-more)
  - [What parameters actually store, from 10B to a trillion](https://edor.ai/en/learn/llm/parameters-what-they-actually-store)
  - [Embeddings, positional encoding and self-attention](https://edor.ai/en/learn/llm/embedding-positional-encoding-self-attention)
  - [Pre-training, self-supervised learning and data cleaning](https://edor.ai/en/learn/llm/pre-training-self-supervised-and-data-cleaning)
  - [Fine-tuning, SFT and where instruction-following comes from](https://edor.ai/en/learn/llm/fine-tuning-sft-and-instruction-following)
  - [Alignment with RLHF and DPO](https://edor.ai/en/learn/llm/alignment-rlhf-and-dpo)
  - [Where hallucination and bias come from, and how to manage them](https://edor.ai/en/learn/llm/where-hallucination-and-bias-come-from)
  - [Knowledge cutoffs and model versioning](https://edor.ai/en/learn/llm/knowledge-cutoff-and-model-versioning)
  - [Hugging Face, the GitHub of AI, and why teachers can browse it too](https://edor.ai/en/learn/llm/hugging-face-the-github-of-ai)
  - [Reading a model card, and what GGUF and Safetensors mean](https://edor.ai/en/learn/llm/model-card-gguf-and-safetensors)
  - [The rise of small language models, from Phi and Gemma to the 8B class](https://edor.ai/en/learn/llm/small-language-models-phi-gemma-and-friends)
  - [Prompt engineering in one lesson, or why the same question gets such different answers](https://edor.ai/en/learn/llm/prompt-engineering-in-one-lesson)
  - [The school knowledge base and retrieval-augmented generation](https://edor.ai/en/learn/llm/knowledge-base-and-rag)
  - [Chunking, reranking and lost in the middle](https://edor.ai/en/learn/llm/chunking-reranking-and-lost-in-the-middle)
  - [Context caching and cost control, or why asking about the same document again is cheaper](https://edor.ai/en/learn/llm/context-caching-and-cost-control)
  - [From chatbot to AI agent, with tool calling and MCP](https://edor.ai/en/learn/llm/ai-agent-tool-calling-and-mcp)
  - [What LangChain and LlamaIndex actually do](https://edor.ai/en/learn/llm/langchain-and-llamaindex)
  - [Parameter-efficient fine-tuning with LoRA and QLoRA on one GPU](https://edor.ai/en/learn/llm/peft-lora-and-qlora-on-one-gpu)
  - [Quantisation, from FP16 down to INT8 and INT4](https://edor.ai/en/learn/llm/quantization-fp16-to-int4)
  - [Running models in school with Ollama and LM Studio](https://edor.ai/en/learn/llm/ollama-and-lm-studio-local-deployment)
  - [On-device AI, Apple Intelligence and the data-never-leaves advantage](https://edor.ai/en/learn/llm/on-device-ai-and-apple-intelligence-privacy)
  - [AI guardrails with Llama Guard, NeMo Guardrails and enforced JSON](https://edor.ai/en/learn/llm/ai-guardrails-llama-guard-and-nemo)
  - [Jailbreaks and prompt injection, and how a school defends against them](https://edor.ai/en/learn/llm/jailbreak-and-prompt-injection-defence)
  - [Privacy, bias and copyright — a school's AI ethics checklist](https://edor.ai/en/learn/llm/privacy-bias-and-copyright-ethics)
  - [Automated evaluation and leaderboards, from MMLU to Chatbot Arena](https://edor.ai/en/learn/llm/llm-as-a-judge-mmlu-gpqa-chatbot-arena)
  - [What comes next — multimodality, reasoning models and embodied AI](https://edor.ai/en/learn/llm/multimodal-reasoning-models-and-embodied-ai)
- [Prompt Engineering Classroom: classroom practice](https://edor.ai/en/learn/prompt-engineering): 18 free lessons, about 233 minutes total. Eighteen lessons of prompt engineering set entirely in Hong Kong classrooms: planning, tiered worksheets, question setting and rubrics, marking feedback, SEN adaptation, the scaffolded tutor, administrative writing, grounded RAG answers, prompt-injection defence and panel template libraries. Every lesson includes copy-ready prompts in both languages.
  - [What makes a good prompt](https://edor.ai/en/learn/prompt-engineering/what-makes-a-good-prompt)
  - [The role, task, context and format framework](https://edor.ai/en/learn/prompt-engineering/role-task-context-format-framework)
  - [Teaching by example, and output consistency](https://edor.ai/en/learn/prompt-engineering/few-shot-examples-and-output-consistency)
  - [Chain of thought and step-by-step instructions](https://edor.ai/en/learn/prompt-engineering/chain-of-thought-and-step-by-step-instructions)
  - [System prompts and school-based personas](https://edor.ai/en/learn/prompt-engineering/system-prompts-and-school-personas)
  - [Lesson planning prompts](https://edor.ai/en/learn/prompt-engineering/lesson-planning-prompts)
  - [Worksheet and differentiation prompts](https://edor.ai/en/learn/prompt-engineering/worksheets-and-differentiation-prompts)
  - [Question setting and rubric prompts](https://edor.ai/en/learn/prompt-engineering/question-setting-and-rubric-prompts)
  - [Marking and feedback prompts](https://edor.ai/en/learn/prompt-engineering/marking-and-feedback-prompts)
  - [Scaffolded tutor prompts that withhold answers](https://edor.ai/en/learn/prompt-engineering/scaffolded-tutor-prompts-that-withhold-answers)
  - [Teaching students prompt literacy](https://edor.ai/en/learn/prompt-engineering/teaching-students-prompt-literacy)
  - [School administration and parent communication prompts](https://edor.ai/en/learn/prompt-engineering/school-administration-and-parent-communication-prompts)
  - [Grounded prompts with citations](https://edor.ai/en/learn/prompt-engineering/grounded-prompts-with-citations-for-rag)
  - [Prompt injection and keeping student-facing content safe](https://edor.ai/en/learn/prompt-engineering/prompt-injection-and-safe-student-content)
  - [Structured output as tables and JSON](https://edor.ai/en/learn/prompt-engineering/structured-output-json-and-tables)
  - [Multimodal prompts for images, handwriting and audio](https://edor.ai/en/learn/prompt-engineering/multimodal-prompts-images-handwriting-and-audio)
  - [Building a panel prompt template library](https://edor.ai/en/learn/prompt-engineering/building-a-panel-prompt-template-library)
  - [A prompt debugging checklist](https://edor.ai/en/learn/prompt-engineering/debugging-prompts-a-checklist)

## FAQ
- Q: What is Edor.ai?
  A: Edor.ai is an AI school-based learning and teaching platform for Hong Kong primary and secondary schools. It combines AI lesson planning, AI assessment and marking, a scaffolded Socratic tutor, subject resource libraries, a student self-directed learning desk, PIE school analytics and SDAS in one secure system where the school owns the data, with consultancy and training aligned to the EDB AI Education Grant.
- Q: Does the AI tutor give students answers?
  A: No. The tutor uses a scaffolded hint ladder: clarifying questions first, then directional hints, worked examples, and only reveals steps at the level a teacher allows. Teachers set the ceiling per task and can watch, pause or flag conversations live.
- Q: Which subjects and languages are supported?
  A: All subjects, with built-in templates for Chinese, English, Mathematics, General Studies / Science, Humanities and STEM. The UI is Traditional Chinese (Hong Kong) with an English toggle; AI output can be Chinese or English per tool; oral practice supports Cantonese, Putonghua and English.
- Q: Where is data stored and can the school export it?
  A: Each school's data lives in an isolated database and file volume, deployable in Hong Kong cloud or a school-designated environment. Schools can export knowledge assets, resources, curriculum, rubrics and student records at any time; we assist migration and deletion at contract end.
- Q: Who owns the data? Is on-premises different from cloud?
  A: The data always belongs to the school — on-premises or cloud makes no difference. Either way, administrators can download everything (JSON / CSV / original files) from the admin panel in one click, free of charge and without limits. That is how confident we are in the service: schools stay because the platform works, not because their data is locked in.
- Q: Can the school switch providers at any time?
  A: Yes. Schools may download all their data and move to another provider at any time, no reason required; at contract end we assist migration and provide a deletion certificate. AI providers (OpenAI / Azure / Anthropic / Poe / Ollama) can also be switched in the admin panel at any time.
- Q: Can we use our own school domain?
  A: Yes. Add one CNAME record in your DNS (e.g. ai.yourschool.edu.hk → edor.ai) and we issue a TLS certificate and enable it. Staff and students log in on the school's domain, and the login page shows the school crest and name. Supported for both cloud and on-premises deployments.
- Q: What is on-premises deployment and what does the school need?
  A: On-premises means the whole platform (Docker containers) is installed on the school's own server or designated private cloud, so data never leaves the school. Typically a Linux host with 4+ cores / 16 GB RAM and outbound access to the AI provider API; we handle installation, updates and monitoring, the school's IT handles backups and network. Features are identical to the cloud edition, including full export at any time.

## Optional
- [RSS](https://edor.ai/blog/rss.xml)
- [Sitemap](https://edor.ai/sitemap.xml)