DeepTutor

DeepTutor

By @timconsidine

Agent-native personalized AI tutoring

0.0.20 upstream 1.7.2 Updated 12 Nov 2026

CloudronVersions.json:

https://communityapps.appx.uk/cloudron-deeptutor/CloudronVersions.json

Tags

education ai productivity

Description

DeepTutor is an agent-native personalized AI tutoring platform. It combines a FastAPI backend with a Next.js frontend to deliver chat, knowledge-base RAG, notebooks, narrated video lessons, Deep Research / Solve / Question, three-layer memory, learning workspaces, immersive reading, a question bank and a task board — all backed by file/SQLite storage in /app/data.

The upstream v1.7.x series is modular: the web app and each feature module are separate distributions. This package installs upstream's Full app profile, so every standard surface is present out of the box.

Document parsing

Knowledge-base ingestion ships with MarkItDown (all format extras) as the default parsing engine, covering PDF, modern and legacy Office, e-mail, audio, images, notebooks, feeds, archives and text — no runtime model download. The built-in text extractor remains as an automatic fallback. Apache Tika is also supported: run a Tika 4.x server yourself and set its URL in Settings → Document Parsing.

Video lessons

Video Lessons generate an animated lesson that plays in the browser. The backend renders with Node.js 24 and npm (both bundled) and downloads its pinned Remotion packages into its runtime state directory under /app/data/user/.runtime/remotion the first time a lesson is rendered. Narration uses the Text-to-Speech provider you configure in Settings, falling back to the bundled edge-tts provider (Microsoft Edge read-aloud voices, no API key) — edge-tts needs no extra tools, while other speech providers additionally require ffmpeg, which this package does not install. All of it lives under /app/data, so it survives restarts and is included in Cloudron backups.

Partner / IM channels

Partner agents can connect to Telegram, WeCom (WeChat Work), Feishu/Lark, DingTalk, Slack, QQ, Zulip, MS Teams and OneBot (NapCat). Connect a channel under Partners → choose a partner → Channels. For Telegram, create a bot with @BotFather, paste its token, and add your Telegram account id (or id|username) to the channel's allowed senders before starting it. Telegram uses outbound long polling, so no inbound port is required.

What this package does NOT include

This is a trimmed Cloudron build. The following upstream optional components are intentionally not installed, so their features are unavailable or fall back to a simpler path. DeepTutor's modules screen lists them as not ready and names the missing package:

  • Matrix (Element) partner channel. The other partner/IM channels — Telegram, WeCom, Feishu/Lark, DingTalk, Slack, QQ, Zulip, MS Teams and OneBot — are included (see above).
  • Math Animator (Manim) and its LaTeX rendering prerequisites.
  • The GraphRAG knowledge-base pipeline (graphrag) — upstream supports it on Python < 3.14 only. The bundled LlamaIndex, PageIndex and BM25 pipelines cover indexing and retrieval instead.
  • LibreOffice-rendered Office previews. Word/PowerPoint/Excel previews fall back to in-browser renderers (DOCX/XLSX) or extracted text.
  • Anything needing ffmpeg: native speech-to-text for browser recordings (DashScope/Volcengine WebM/Opus), transcribing uploaded audio/video under Reading, joining speech segments in Audio Overview, and video-lesson narration through a non-Edge Text-to-Speech provider. Narration falls back to the bundled edge-tts voices, which need no ffmpeg.
  • Extra document-parsing engines (Docling, PyMuPDF4LLM, LiteParse, local MinerU) and a bundled Tika server. MarkItDown is included; MinerU can be used in cloud mode with an API token.
  • Re-ranking (sentence-transformers/torch), CodeBuddy in-app login, and local OCR of scanned PDFs.
  • External sandbox-runner / PocketBase / Redis sidecars. In-container restricted-subprocess execution and in-memory coordination are used instead.
  • Codex OAuth sign-in (needs loopback ports Cloudron does not expose); configure Codex with an API key instead.

Core chat, knowledge bases, research/solve/question, notebooks, video lessons, memory, reading and the task board work without these extras.

***

Packaged by @timconsidine https://communityapps.appx.uk (full portfolio: https://myca.appx.uk)

Screenshots

Recent Changes

* 0.0.20 : upstream 1.7.2