Introduction

Integrating models and running Agents in DeepSeek Harness (DSH) addresses the ability to “ask questions, write code, and modify code.” However, to systematically learn a new topic, the common practice remains self-structuring textbooks, taking notes, or treating long conversations as a classroom setting—structures are often loose, and it is difficult to visualize derivations while lecturing. The client plugin Aha Tutor (dsh-aha-tutor), open-sourced by community maintainer BaronCyrus, consolidates this path: it first assesses your background and goals to generate an editable course blueprint, then proceeds step-by-step through lectures on a handwritten blackboard with voice and subtitles, accompanied by quizzes, exercises, and reviews. The following sections introduce what it is, what it does, and how to install and get started.

What is this?

Aha Tutor · AI Private Tutor (npm package name dsh-aha-tutor) is a plugin for the DSH Desktop client, categorized as “Client”. In a nutshell: it equips DeepSeek Harness with a private tutor—state the content you want to master (or attach teaching materials), and the plugin first assesses you, then customizes a “Unit → Chapter → Lecture” curriculum, teaching step-by-step on a handwritten blackboard.

The plugin is not affiliated with DeepSeek or any other learning products. The DSH ecosystem emphasizes “Everything is a plugin,” and the SkillHub directory is a community site, not an official app store.

Core Features

The following capabilities are based on the project’s README.

Customized Courses and Material Import

State what you want to learn, or upload teaching materials; the tutor can first search the web for the topic, then use multiple-choice questions to understand your background, goals, available time, and course scale, while writing out the course blueprint on the blackboard. The blueprint can be directly edited, split into lectures, or deepened; you can also suggest modifications in a single sentence before applying them. Upon confirmation, it generates a practical project and exercises for the first unit; exercises for subsequent lectures are provided when you reach them.

Materials support PDF, Word, PPT, Markdown, web links, and images (image recognition requires a model with vision capabilities). Lectures and quick reference sheets can cite original page numbers, and long documents can be “digested” into structures, arguments, and glossaries.

Handwritten Blackboard Classroom

Voice explanations are paired with subtitles; key points, formulas, diagrams, and charts appear on the blackboard in handwriting, word by word, with the current sentence highlighted by a highlighter. The entire lecture’s board work is on a draggable, zoomable canvas, with the camera following the explanation; a sidebar shows key point progress and lecture records. It supports pause, replay, and playback speed adjustment.

In-class Quizzes, Q&A, and Adaptive Learning

The tutor pauses at critical points to present questions and asks “How confident are you?”; guessing correctly, hesitation, and confident wrong answers are treated differently. Lectures involving calculations can request written processes, and corrections are made on the blackboard if errors occur. After finishing a lecture, weak knowledge points can trigger automatic reinforcement lessons. You can “raise your hand” to ask questions (via text or DSH native voice input) or request a different explanation on the blackboard (using analogies, drawings, examples, or adjusting depth).

Exercises, Flashcards, Planning, and Export

Each lecture includes exercises (possibly with diagrams; short-answer questions are graded by AI). Flashcards and knowledge points are scheduled for review using spaced repetition, and the study plan reserves time for reviews. Printable quick reference sheets, notes, flashcard sets, and quizzes can be generated by topic.

Inputting a topic or URL generates a roughly two-minute whiteboard explanation short video; individual lectures can be exported as videos with board work animation, subtitles, and watermarks. Board work can be exported as PDF (one page per image, with handwritten fonts and formulas) or as page-by-page PNGs.

Practical Projects and DSH Conversations

Code-based practical projects can generate project directories and starter files with one click. Opening a session in DSH allows you to complete the project with a coding Agent, which provides feedback based on scoring criteria upon submission. The plugin also provides a conversation preset called Aha Tutor · AI Private Tutor, allowing you to use whiteboard explanations, interactive quizzes, course building, or quick reference sheets in regular conversations.

The home page features built-in sample courses like “Simple Harmonic Motion” and “Bayesian Thinking”, allowing you to experience the full blackboard classroom without creating a course first. The home page also supports quick access entries like “Explain a Concept,” “Help me Solve a Problem,” and “Draw it for me.”

Models and Voice

It defaults to using the default model configured in DSH; in Aha Tutor → Settings, you can specify models separately for “Course Creation & Lecturing” and “Grading & Q&A”. The README states: The course structure is generated upon confirmation, each lecture’s board work is generated when that lecture is opened (and the next lecture is prepared), and exercises are also generated upon opening. It does not consume quota in batches in the background.

Voice uses macOS / Windows system TTS (free) by default; cloud voices (OpenAI-compatible interfaces, Volcano Engine, MiniMax, Azure, etc.) are optional. Keys are stored in the DSH credential vault and used only on the Host side. When exporting videos, only cloud voices can be recorded; system voice exports result in videos with subtitles but no voiceover.

Installation and Activation

Prerequisites: DeepSeek Harness Desktop version 0.2.x (0.2.1 or higher), and at least one available model configured in DSH. If DSH is not installed, please refer to DSH official instructions.

  1. Open the Plugins page in the desktop app, click Add Plugin, and fill in the Package Name or Address:
dsh-aha-tutor

Select Official Registry for the installation source and click Install. If a new version has been released for less than 24 hours, DSH’s pnpm might delay installation due to release cooling periods. You can specify an exact version, for example:

dsh-aha-tutor@1.5.0

  1. Completely quit and reopen DSH (refreshing the page alone will not load new Host-side plugins).

  2. The Aha Tutor · AI Private Tutor entry will appear in the left sidebar. Click it to start using.

Offline installation: You can download dsh-aha-tutor-<version>.tgz from GitHub Releases and fill in the absolute path to the file in the package name or address field. For self-created profiles, you can use dsh plugin --profile <name> add dsh-aha-tutor@<version>; do not use this command for the desktop profile—use the plugin page for the desktop version.

Updates: Check for updates in Aha Tutor → Settings → Plugin Version; after updating, you must also completely quit and restart DSH. You can disable or uninstall the plugin in the Plugins page. Uninstalling does not delete local learning data directories.

Typical Usage

The README’s “Get Started in 3 Minutes” flow can be reproduced directly:

  1. State Your Goal: Enter something like “Understand eigenvalues in linear algebra within three weeks” on the home page, or attach a handout, and click Start.
  2. Answer Questions: Select your background, purpose, and pace, then click Generate Course Blueprint. If you are not satisfied, you can edit it manually or write a sentence like “add more programming exercises.”
  3. Confirm and Start: Click Confirm, Generate Entire Course. On the course page, click Continue Learning, and the tutor begins lecturing on the blackboard.

To use the tutor within DSH: When creating a new session, select the preset Aha Tutor · AI Private Tutor.

Data Storage and Privacy

Courses, progress, materials, profile, and study plans are stored locally:

$DSH_HOME/aha-tutor/

The default is ~/.dsh/aha-tutor/ (Windows: C:\Users\<Username>\.dsh\aha-tutor\). Practical project directories are in ~/AhaTutor/projects/; the plugin will not overwrite files you have already modified. Content sent to models consists of fragments necessary to complete tasks (questions, course structure, relevant materials, and brief learning profiles). Web import only fetches public addresses and does not access local or LAN addresses. For details, see the repository’s SECURITY.md.

Use Cases and Notes

This is suitable for users who wish to systematically learn a topic within the local DSH environment, requiring visualized board work and pace control (pause, replay, speed adjustment). It is also suitable for learners who want to convert PDFs/handouts into structured courses or need flashcards, planning, and video/PDF exports. Developers can run pnpm install --frozen-lockfile and pnpm run check in the repository for local verification; Host code changes require restarting DSH.

Before installing any DSH plugin, it is recommended to review the source code and MIT license yourself, and understand that the plugin runs with the permissions of the current DSH process. Aha Tutor relies on your configured models and optional cloud voice keys; please comply with the corresponding service terms when using subscription-based models.

Conclusion

Aha Tutor consolidates “Customized Curriculum + Blackboard Lectures + Test/Practice/Review” into a DSH client plugin, with data stored locally by default. For further details, please refer to the repository’s Usage Guide (Chinese).