Skip to main content
Guided Learning converts the records saved in your notebooks into a structured, progressive learning experience. A pipeline of four specialized agents analyses your material, generates an interactive HTML lesson for each knowledge point, answers questions in context, and summarizes what you have learned.

​
Multi-agent flow

Notebook records
      ↓
 LocateAgent        — identifies 3–5 progressive knowledge points
      ↓
 InteractiveAgent   — generates an interactive HTML page per knowledge point
      ↓
 ChatAgent          — answers questions during the lesson (per knowledge point)
      ↓
 SummaryAgent       — generates a personalized learning summary at the end

​
Agent responsibilities

AgentResponsibility
LocateAgentAnalyses all records in the selected notebook(s) and organizes them into 3–5 core knowledge points arranged in a progressive learning order
InteractiveAgentConverts each knowledge point into a visual, interactive HTML page with appropriate interactive elements based on the content type; includes HTML bug-fixing capability
ChatAgentAnswers questions during a knowledge point lesson; responses are grounded in the current knowledge point and your chat history within that point
SummaryAgentAfter all knowledge points are completed, generates a learning summary covering what was covered, mastery assessment, and follow-up suggestions

​
Starting a guided learning session

1

Open guided learning

Navigate to http://localhost:3782/guide (or access it from the dashboard).
2

Select one or more notebooks

Choose the notebook(s) that contain the records you want to learn from. Cross-notebook selection is supported — you can combine records from multiple notebooks into a single session.
3

Generate the learning plan

Click Generate Learning Plan. LocateAgent analyses the notebook content and identifies 3–5 knowledge points in progressive order. The plan is displayed before you start so you can see what will be covered.
4

Start learning

Click Start Learning. InteractiveAgent generates the HTML page for the first knowledge point. The interactive page appears on the right side of the screen.
5

Ask questions

Type questions in the chat box on the left. ChatAgent answers based on the current knowledge point and your conversation history within it.
6

Move to the next knowledge point

When you are ready, click Next. InteractiveAgent generates the page for the following knowledge point.
7

Complete the session

After the final knowledge point, SummaryAgent generates your personalized learning summary, including a mastery assessment and recommendations for further study.

​
Cross-notebook selection

You can draw records from multiple notebooks into a single guided learning session. This is useful when you have notes spread across different notebooks — for example, solve results in one notebook and research summaries in another — and want a unified lesson that connects them. Select multiple notebooks on the session setup screen before generating the learning plan. LocateAgent merges all records from the selected notebooks when identifying knowledge points.

​
Interactive HTML pages

Each knowledge point is rendered as a self-contained HTML page designed to make the concept concrete and engaging. InteractiveAgent chooses the interaction style based on the content:
  • Step-by-step breakdowns for algorithms and proofs
  • Diagrams and visual aids for conceptual material
  • Worked examples for mathematical content
If the LLM generates HTML with rendering issues, the system includes a built-in debug and fix step. You can also trigger an HTML fix manually via the API endpoint POST /api/v1/guide/fix_html.
Interactive pages are generated on demand — each page is created when you advance to that knowledge point, not upfront. Generation takes a few seconds.

​
Session persistence

Session state is saved automatically throughout the session. This includes:
  • The full knowledge point plan
  • The HTML page for each visited knowledge point
  • The complete chat history for each knowledge point
  • Your current position in the learning plan
If you close the browser or lose connectivity, you can resume exactly where you left off by returning to the same session. Sessions are stored indefinitely until you delete them. 
data/user/guide/
└── session_{session_id}.json    # Complete session state, knowledge points, chat history

​
API reference

The guided learning module exposes a REST API and a WebSocket endpoint for real-time interaction.
MethodEndpointDescription
POST/api/v1/guide/create_sessionCreate a new learning session
POST/api/v1/guide/startStart learning (triggers first InteractiveAgent run)
POST/api/v1/guide/nextAdvance to the next knowledge point
POST/api/v1/guide/chatSend a chat message to ChatAgent
POST/api/v1/guide/fix_htmlRe-run HTML generation for the current page
GET/api/v1/guide/session/{session_id}Retrieve full session state
GET/api/v1/guide/session/{session_id}/htmlRetrieve the current HTML page
Connect to WS /api/v1/guide/ws/{session_id} for real-time streaming of agent outputs and progress events during HTML generation and chat responses.

​
Tips for best results

Guided Learning works best when your notebooks contain rich records from multiple DeepTutor modules — solve results, research summaries, and generated questions all contribute context that LocateAgent uses to build a well-rounded learning plan.
The system automatically generates 3–5 knowledge points. If a notebook has very few records, the plan may be shorter. Add more records to the notebook first for a more comprehensive session.

Build docs developers (and LLMs) love

Get started for freeTalk to us