AI Chat Features

SnapThink’s AI chat is powered by local language models, giving you intelligent assistance while keeping everything private.

How It Works

SnapThink uses Ollama to run AI models directly on your computer:

• No internet required once models are downloaded
• Complete privacy - conversations never leave your device
• Multiple model support - choose the best model for your task

Available Models

Recommended Models

Llama 3.2 (3B) - Best for most users

ollama pull llama3.2:3b

• Use for: General conversation, document Q&A, basic coding
• RAM needed: 4-6GB
• Speed: Fast responses

Llama 3.2 (8B) - More powerful

ollama pull llama3.2:8b

• Use for: Complex analysis, detailed explanations, advanced coding
• RAM needed: 8-12GB
• Speed: Slower but more capable

CodeLlama (7B) - Coding specialist

ollama pull codellama:7b

• Use for: Python programming, code review, debugging
• RAM needed: 6-8GB
• Speed: Medium, optimized for code

Specialized Models

Mistral (7B) - Balanced performance

ollama pull mistral:7b

• Good balance of speed and capability
• Excellent for reasoning tasks

Qwen2.5 (7B) - Multilingual support

ollama pull qwen2.5:7b

• Strong multilingual capabilities
• Good for international documents

Chat Capabilities

🤖 General Conversation

Ask anything you’d ask ChatGPT:

• Explain complex topics
• Get writing assistance
• Brainstorm ideas
• Answer factual questions

Example:

User: Explain machine learning to a 10-year-old
AI: Imagine teaching a computer to recognize cats in photos...

📄 Document-Aware Chat

When you upload documents, the AI can:

• Answer questions about document content
• Summarize key points
• Find specific information
• Compare information across documents

Example:

User: What are the main conclusions in my research paper?
AI: Based on your uploaded paper, the main conclusions are:
1. The proposed method improves accuracy by 15%...

🐍 Python Code Execution

The AI can write and run Python code:

• Data analysis and visualization
• Mathematical calculations
• File processing
• Statistical analysis

Example:

User: Calculate the correlation between age and income in my dataset
AI: I'll analyze the correlation for you:

[Python code block]
import pandas as pd
import numpy as np

# Load your dataset
df = pd.read_csv("your_data.csv")
correlation = df['age'].corr(df['income'])
print(f"Correlation: {correlation:.3f}")

[Result: Correlation: 0.742]

Advanced Features

Context Memory

The AI remembers your entire conversation:

• Reference previous questions and answers
• Build on earlier analysis
• Maintain context across the session

Multi-Document Analysis

Upload multiple documents and ask comparative questions:

• “Compare the findings in paper A vs paper B”
• “What themes appear across all my documents?”
• “Find contradictions between these sources”

Code Persistence

Python variables and imports persist throughout the conversation:

• Load data once, use it multiple times
• Build complex analysis step by step
• Create and reuse custom functions

Best Practices

💡 Writing Effective Prompts

Be Specific

❌ Vague: “Analyze my data” ✅ Specific: “Create a histogram of customer ages and identify the most common age group”

Provide Context

❌ No context: “What does this mean?” ✅ With context: “In the context of my sales report, what does the 15% increase in Q3 indicate?”

Break Down Complex Tasks

❌ Too complex: “Analyze everything and create a complete report” ✅ Step by step:

1. “First, show me summary statistics for my dataset”
2. “Now create visualizations for the top 3 variables”
3. “Finally, identify any correlations or patterns”

🎯 Getting Better Results

Use Follow-up Questions

• “Can you explain that in simpler terms?”
• “What are the implications of this finding?”
• “Show me a different type of visualization”

Request Specific Formats

• “Summarize this in bullet points”
• “Create a table comparing these options”
• “Write this as a formal conclusion”

Ask for Code Comments

• “Add comments to explain the code”
• “Show me step-by-step what this code does”
• “Explain the logic behind this approach”

Model Selection Guide

When to Use Each Model

Llama 3.2:3b - Daily driver

• ✅ Quick questions and answers
• ✅ Document summarization
• ✅ Simple data analysis
• ✅ Basic coding tasks
• ❌ Complex reasoning
• ❌ Very long documents

Llama 3.2:8b - Heavy lifting

• ✅ Complex analysis tasks
• ✅ Detailed explanations
• ✅ Advanced coding
• ✅ Large document processing
• ❌ When you need speed
• ❌ On low-RAM systems

CodeLlama:7b - Programming focus

• ✅ Writing Python scripts
• ✅ Code debugging
• ✅ Algorithm explanations
• ✅ Data science workflows
• ❌ General conversation
• ❌ Non-coding tasks

Switching Models

1. Click the model name in the top bar
2. Select from your downloaded models
3. The conversation context carries over

Performance Tips

Speed Optimization

• Use smaller models for quick tasks
• Close other applications to free up RAM
• Use GPU acceleration if available

Quality Optimization

• Use larger models for complex tasks
• Provide clear, detailed prompts
• Break complex requests into steps

Memory Management

• Restart conversations if they get very long
• Create new notebooks for different projects
• Monitor system RAM usage

Troubleshooting

Common Issues

“No models available”

1. Install Ollama: curl -fsSL https://ollama.com/install.sh | sh
2. Download a model: ollama pull llama3.2:3b
3. Restart SnapThink

Slow responses

• Try a smaller model (3b instead of 8b)
• Close other applications
• Check available RAM

Poor quality answers

• Use a larger model
• Provide more context in your prompts
• Break complex questions into parts

Model won’t download

• Check internet connection
• Ensure sufficient disk space
• Try downloading manually: ollama pull model-name

Next Steps

• Document Analysis - Learn to work with uploaded files
• Python Environment - Master the coding capabilities
• CSV Analysis Guide - Practical data analysis examples

Pro tip: Start with simple questions to get familiar with your model’s style, then gradually work up to more complex tasks!