UniMind Engine

UniMind is UniFlow's proprietary AI reasoning engine. It goes far beyond a simple chatbot — it builds a vector knowledge graph for every course you take, enabling deep retrieval across 8 data source types. Every answer UniMind gives is grounded in your actual course materials, with citations you can verify.

How UniMind Works

1. Data Ingestion

When you sync a course from Canvas, UniFlow extracts content from:

2. Smart Chunking & Embedding

Course content is processed through a smart chunking pipeline:

1. Content is broken into semantically meaningful chunks
2. Each chunk is deduplicated by content hash
3. Chunks are sent to a cloud Edge Function for vector embedding
4. Embedded vectors are stored in your personal knowledge graph

This happens automatically in the background after each Canvas sync.

3. Retrieval & Reasoning

When you ask a question:

1. Task Router analyzes your query to determine the best approach (Q&A, coverage analysis, or planning)
2. Vector search retrieves the most relevant chunks from your knowledge graph
3. Course Map provides structural context (what modules exist, which assignments are upcoming)
4. AI model generates an answer using the retrieved context, with citations

Using UniMind

Opening UniMind

• Click UniMind in the sidebar
• Or press Cmd+/ (Mac) / Ctrl+/ (Windows) from anywhere

Scope Selection

Before asking a question, select your scope:

• Single Course — Focus on one specific course's materials
• All Courses — Search across all your synced courses

The scope determines which knowledge graph is searched.

Asking Questions

Type your question in the input field and press Enter. Examples:

• "What are the key themes in Chapter 5?"
• "When is the midterm and what does it cover?"
• "Explain the difference between X and Y from the lecture notes"
• "Help me outline an essay for Assignment 3"
• "What's the grading breakdown for this course?"

Citations

UniMind's answers include citations — references to the specific course materials used to generate the response. Each citation shows:

• The source type (assignment, module, announcement, etc.)
• The relevant text excerpt
• A clickable link to view the full source

This lets you verify every claim the AI makes against your actual course data.

Task Types

UniMind automatically detects what kind of help you need:

Q&A Mode

For specific questions with concrete answers:

• "What is the due date for Essay 2?"
• "Explain the concept of X from the lecture"
• "What are the submission requirements?"

Coverage Analysis

For comprehensive overviews and study preparation:

• "Give me an overview of everything we've covered so far"
• "What topics should I review for the final?"
• "Summarize all the key concepts in this course"

Coverage mode provides a structured report with evidence from multiple source types.

Planning Mode

For study plans and task organization:

• "Create a study plan for the final exam"
• "What should I prioritize this week?"
• "Help me plan my approach for the research paper"

Advanced Features

Web Search Integration

When your course materials don't contain enough context, UniMind can optionally search the web:

• Toggle web search on/off in the chat interface
• Web sources are displayed separately from course citations
• Search queries and quality metrics are shown for transparency

Canvas Content Generation

UniMind can generate formatted content (essays, outlines, responses) in a canvas panel:

1. Ask UniMind to write or draft something
2. Content appears in a side canvas with rich formatting
3. Click Open in Editor to transfer it to the Document Editor
4. Continue editing with full formatting tools

Streaming Responses

All responses stream in real time — you see the answer being generated word by word, so you don't have to wait for the complete response.

Conversation History

Each chat session maintains full conversation history:

• Multi-turn conversations with context retention
• Smart history management with token budgeting
• Create new sessions for different topics
• Switch between sessions in the sidebar

Assignment-Aware Context

When you ask about a specific assignment, UniMind automatically:

1. Loads the full assignment description, rubric, and attachments
2. Classifies the assignment type (essay, discussion, quiz, etc.)
3. Retrieves peer exemplars if available
4. Generates contextually appropriate responses

AI Models

UniMind supports multiple AI providers:

• Google Gemini — High-quality reasoning with large context windows
• OpenAI GPT — Including GPT-4 for complex tasks

The specific model used depends on your plan and admin configuration.

Usage Limits

When you reach your limit, a paywall prompt appears with upgrade options. Usage resets monthly.

Tips

• Be specific — The more specific your question, the better UniMind can retrieve relevant context
• Select the right scope — Use single-course scope when asking about a specific class, all-courses scope for cross-course questions
• Check citations — Always verify important claims by clicking the citation links
• Use for exam prep — Ask UniMind to create study guides, practice questions, or topic summaries
• Generate, then edit — Use UniMind to draft content, then refine it in the Document Editor

Troubleshooting

"No relevant context found"

This means UniMind couldn't find matching content in your knowledge graph. Try:

1. Ensuring your course is fully synced (deep scan)
2. Rephrasing your question with different keywords
3. Checking that the course index has been built (visible in the scope panel)

AI service not configured

This typically means the backend AI configuration hasn't been set up. Contact support if you see this error.

Slow responses

Response time depends on the complexity of your question and the amount of context retrieved. Coverage analysis queries take longer than simple Q&A.

What's Next

• Canvas Sync — Ensure your courses are fully synced for the best AI experience
• Live Transcription — Add lecture transcripts to your knowledge graph
• Writing Assistant — Polish AI-generated content
• AI Shield — Check your writing for AI detection before submitting