TeXposit Tutorial

16. AI Agent Settings & Configuration

Configure how AI assists your writing and reasoning.

5 min read Share

The AI assistant works out of the box, but the settings give you control over response speed, reasoning depth, and task structure.

AI Model Selection

TeXposit offers three model tiers, each optimized for different use cases:

Fast (Default)

Lightning-quick responses perfect for:

  • Brainstorming and drafting
  • Quick LaTeX syntax questions
  • Simple formatting requests

Tradeoff: Less capable for complex reasoning or restructuring.

General (Recommended)

Sweet spot of speed and capability:

  • Most writing improvement tasks
  • Inline suggestions and corrections
  • Outline generation and restructuring
  • Reference and citation assistance

Best For: Most users, most tasks. Default recommendation.

Smart (Maximum Quality)

Most capable model for complex tasks:

  • Deep reasoning and complex restructuring
  • Multi-step task planning
  • Advanced writing refinement
  • Understanding nuanced research context

Tradeoff: Slower responses. Best for important revisions.

Planning Mode

Control how the AI approaches complex tasks:

Auto (Default)

TeXposit decides whether to use task sessions based on complexity. Multi-step requests get a structured session; simpler ones get a direct response. Best for: most users.

Planning

Always use task sessions for complex requests. The AI breaks work into steps, shows you a plan, and executes systematically. Best For: Users who want to see and approve AI's plan before execution.

Off

Never use task sessions. All requests are handled in direct conversation mode. Best For: Quick iterations and conversational back-and-forth.

Context & Awareness

The AI assistant automatically analyzes your project to provide better suggestions:

Bibliography Analysis

Reads your .bib file to suggest citations and maintain consistency with your bibliography style.

Document Structure

Understands your section hierarchy to maintain narrative flow and avoid duplicate content.

Terminology

Learns the key terms and vocabulary you use to maintain consistency throughout your document.

Open Files

Considers what you're currently viewing to provide localized and relevant suggestions.

Using Local Models (Ollama)

For privacy or offline work, you can configure TeXposit to use local models running on your machine:

  1. Install and run Ollama locally (see tutorial 12)
  2. Start a model in Ollama (e.g., ollama run qwen2.5-coder)
  3. In TeXposit, enable "Use Local Ollama" in the AI settings
  4. Select the model you want to use
  5. Start using the AI assistant. It will now use your local model.

Privacy Note: When using local Ollama models, all your text stays on your machine. Nothing is sent to cloud servers. Perfect for sensitive research or working offline.

Suggestion Behavior

Control how AI suggestions appear when you request inline improvements:

Always Review (Default)

All AI suggestions appear as editable proposals. You review each one and decide to accept or reject.

Auto-Accept

Non-controversial suggestions (like fixing typos or formatting) are applied automatically. Complex suggestions still show for review. Use this if you trust the AI and want faster iterations.

Disabled

Turn off inline AI suggestions completely. You can still use the AI chat directly, but no automatic suggestions will appear in the margin.

Conversation History

The AI assistant maintains context across messages:

  • Persistent Memory: The AI remembers your conversation within the current session, so follow-up requests are more coherent
  • Context Window: Each conversation has a limited context window. Very long chats may lose early messages.
  • New Session: Starting a new chat clears history. Useful if you want a fresh start on a different topic

Recommended Settings by Use Case

📝 Drafting Mode

Model: Fast | Planning: Off | Suggestions: Auto-Accept

Fast iteration while drafting.

✏️ Editing Mode

Model: General | Planning: Auto | Suggestions: Always Review

Refining existing content with review control.

🎯 Complex Restructuring

Model: Smart | Planning: Planning | Suggestions: Always Review

Major revisions. See the plan first and review all changes.

🔒 Privacy-First

Local Ollama | Model: Qwen2.5-Coder 7B | Planning: Auto

Everything stays on your machine. Good for sensitive research.

Pro Tips

Start with General: Most users shouldn't need Fast or Smart. General is the ideal balance.

Use Planning Mode for Projects: When working on longer documents, Planning Mode helps the AI break work into manageable chunks.

Switch Models Mid-Work: Start with Fast for drafting, then switch to Smart for final polish. You can change settings anytime.

Test Local Models: Try Ollama with a 7B model on your system. Many users are surprised how capable they are.