Skip to content

Chat Interface Guide

Ollama PDF RAG offers two user interfaces: a modern Next.js app and a classic Streamlit interface. This guide covers both.

The Next.js interface at http://localhost:3000 provides the full-featured experience.

Next.js Chat Interface

Layout

┌────────────────────────────────────────────────────────────────┐
│  PDF Chat                                    [GitHub] [Theme]  │
├──────────────────┬─────────────────────────────────────────────┤
│                  │                                             │
│  📄 Documents    │         Chat Messages                       │
│  (1/2 selected)  │                                             │
│  ────────────────│         ┌─────────────────────────────┐    │
│                  │         │ User: What is this about?    │    │
│  ☑️ Policy.pdf   │         └─────────────────────────────┘    │
│    45 chunks     │                                             │
│                  │         ┌─────────────────────────────┐    │
│  ☐ Guide.pdf    │         │ Assistant: Based on the...   │    │
│    23 chunks     │         │ Sources: Policy.pdf (3,7)    │    │
│                  │         └─────────────────────────────┘    │
│  ────────────────│                                             │
│                  │                                             │
│  Today           │                                             │
│  ○ Chat 1        │  ┌──────────────────────────────────────┐  │
│  ○ Chat 2        │  │ Send a message...      📎 qwen3:8b ↑ │  │
│                  │  └──────────────────────────────────────┘  │
└──────────────────┴─────────────────────────────────────────────┘

Components

The sidebar shows all uploaded PDFs with checkboxes:

📄 Documents (2/3)
⚠️ Select PDFs to use as context

☑️ Security_Guide.pdf
   45 chunks • 12 pages

☑️ Policy_Manual.pdf
   23 chunks • 8 pages

☐ Reference.pdf
   15 chunks • 4 pages

[All] [None]          🗑️
Element Description
Checkbox (☑️/☐) Toggle PDF for RAG context
Chunk count Number of text segments
Page count Original PDF pages
All/None Quick select/deselect all
🗑️ Delete PDF (on hover)

Selection Persists

Your PDF selection is saved to localStorage and persists across browser sessions.

Chat History

Shows previous conversations grouped by date:

  • Today - Chats from today
  • Yesterday - Previous day
  • Last 7 days - Past week
  • Older - Everything else

Click any chat to resume the conversation.

Chat Input

┌─────────────────────────────────────────────────────┐
│ Send a message...                                    │
│                                                      │
│ 📎                         ⚙️ qwen3:8b            ↑ │
└─────────────────────────────────────────────────────┘
Button Action
📎 Upload PDF file
⚙️ Model Select Ollama model
↑ Send message

Features

1. PDF Upload

Click 📎 → Select PDF → Wait for processing

Processing includes: 1. Save file to server 2. Extract text (UnstructuredPDFLoader) 3. Split into chunks (7500 chars) 4. Generate embeddings (nomic-embed-text) 5. Store in ChromaDB

2. Document Selection

Pre-chat Selection: Select PDFs BEFORE sending your first message.

✅ Correct workflow:
1. Upload PDFs
2. Select relevant PDFs (☑️)
3. Ask question

❌ Won't work well:
1. Upload PDFs
2. Ask question immediately
3. (No PDFs selected!)

3. Question Classification

The system automatically detects your intent:

Question Type Detection Action
Document query "what does the document say..." Uses RAG
General chat "what is machine learning?" Direct LLM
Doc query, no selection Document keywords, no PDFs Shows warning

Warning message when no PDFs selected: 

⚠️ No documents selected

It looks like your question might be about a document,
but you haven't selected any PDFs to search.

To get answers from your documents:
1. Look at the sidebar on the left
2. Check the boxes next to the PDFs you want to use
3. Then ask your question again

4. Model Selection

Switch between available Ollama models:

Model Size Best For
llama3.2 2GB Fast responses
qwen3:8b 5GB Deep reasoning
mistral 4GB Balanced
deepseek-r1 4GB Complex analysis

Models with "thinking" support (qwen3, deepseek-r1) show their reasoning process.

5. Chat Persistence

  • Chats auto-save to SQLite database
  • Resume any previous conversation
  • Delete individual chats or all history
  • Chat titles auto-generated from first message

6. Response Streaming

Responses stream word-by-word with: - Reasoning steps (for thinking models) - Main answer text - Source citations

Streamlit Interface (Classic)

The Streamlit interface at http://localhost:8501 provides a simpler experience.

Streamlit Interface

Layout

┌─────────────────────────────────────────────────────────┐
│  Ollama PDF RAG                                         │
├───────────────┬─────────────────────────────────────────┤
│               │                                         │
│  📄 Upload    │              PDF Preview                │
│  [Browse]     │         ┌──────────────────┐           │
│               │         │                  │           │
│  🤖 Model     │         │   Page 1 of 10   │           │
│  [llama3.2 ▼] │         │                  │           │
│               │         └──────────────────┘           │
│  🔍 Zoom      │                                         │
│  [────●────]  │                                         │
│               ├─────────────────────────────────────────│
│  ❌ Delete    │                                         │
│               │              Chat Area                  │
│               │                                         │
│               │  User: What is this document about?    │
│               │                                         │
│               │  Assistant: This document covers...    │
│               │                                         │
│               │  [Type your question here...]          │
│               │                                         │
└───────────────┴─────────────────────────────────────────┘

Features

Feature Description
File Upload Drag & drop or browse
Sample PDF Quick start with included samples
Model Selection Dropdown of available models
PDF Viewer Preview with zoom control
Chat History In-session message history
Delete Collection Clear vector database

Usage

  1. Upload PDF - Use sidebar uploader or sample
  2. Select Model - Choose from dropdown
  3. Adjust Zoom - Slider for PDF visibility
  4. Ask Questions - Type in chat input
  5. Clear Context - Delete Collection when switching PDFs

Comparison

Feature Next.js Streamlit
Modern UI ✅ ❌
Chat persistence ✅ ❌
Multi-PDF selection ✅ ❌
Question classification ✅ ❌
PDF preview ❌ ✅
Response streaming ✅ ✅
Mobile friendly ✅ ⚠️
Setup complexity Medium Low

Keyboard Shortcuts

Next.js

Shortcut Action
Enter Send message
Shift + Enter New line in message

Streamlit

Shortcut Action
Enter Send message
Ctrl + K Clear chat

Best Practices

For Best Results

  1. Select specific PDFs - Don't use "All" unless needed
  2. Ask focused questions - One topic at a time
  3. Use thinking models - For complex analysis
  4. Check sources - Verify which chunks were used

Common Patterns

# Summary request
"Summarize the key points in this document"

# Specific lookup
"What does section 3.2 say about authentication?"

# Comparison
"How does chapter 1 compare to chapter 5?"

# Extraction
"List all the dates mentioned in this document"

Troubleshooting

Issue Solution
Slow responses Use smaller model, fewer PDFs
Wrong sources Be more specific in question
Missing context Select more PDFs
No response Check Ollama is running