Getting Best Results from Astronox

Tips, techniques, and strategies for optimal performance.

Core Principles

🎯 Be Specific, Not Vague

Vague requests lead to generic responses:

❌ "Find files"
   → AI: "Which files? Where?"

✅ "Find PDF files containing 'invoice' in my Documents folder from last month"
   → AI: [Precise results]

Why it works:

• AI knows exactly what you want
• No back-and-forth clarification
• Faster results
• Better accuracy

🔗 Provide Context

Context helps AI understand your goals:

❌ "This doesn't work"
   → AI: "What doesn't work?"

✅ "This Python script throws 'ModuleNotFoundError' when I run it.
    I installed the package but the error persists."
   → AI: [Targeted solutions]

Include:

• What you're trying to do
• What's happened so far
• What the error is (exact text)
• Environment details (if relevant)

💬 Use Natural Language

Don't write commands, speak naturally:

Good examples:
✅ "Show me which apps are using the most memory"
✅ "Find photos from last summer"
✅ "Help me organize my project files"

AI understands:

• Colloquialisms
• Varied phrasing
• Questions or commands
• Follow-up context

🔁 Iterate and Refine

Start broad, then narrow:

You: "Find large files"
AI: [Shows files >1GB]

You: "Only videos"
AI: [Filters to videos]

You: "From this year"
AI: [Further filtered]

You: "Delete ones I haven't watched"
AI: [Checks access times, suggests deletion]

Benefits:

• Review at each step
• Adjust as needed
• Maintain control
• Learn what works

Writing Effective Prompts

Structure for Success

Template:

[Action] + [Target] + [Location] + [Criteria] + [Desired outcome]

Example:
"Find PDF files in my Documents folder larger than 10MB and sort by date"

• Action: Find
• Target: PDF files
• Location: Documents folder
• Criteria: >10MB
• Outcome: Sorted by date

Use Examples

Show, don't just tell:

❌ "Rename files with a pattern"

✅ "Rename files to include date:
    report.pdf → 2025-12-20-report.pdf
    image.png → 2025-12-20-image.png"

AI sees exact format you want

Break Down Complex Tasks

Instead of one massive request:

❌ "Set up my entire development environment with all tools and projects"

Try sequential steps:

✅ "First, show me what dev tools I already have installed"
   [Review]
✅ "Now install Node.js and Python if missing"
   [Verify]
✅ "Set up my project folder structure"
   [Confirm structure]
✅ "Clone my repositories from GitHub"

Ask for Previews

Before destructive operations:

✅ "Show me which files will be deleted before you delete them"
✅ "List the changes you'll make first"
✅ "Preview the script you'll generate"

Maximizing AI Capabilities

Leverage Memory

Store frequently used info:

Once: "Remember my project folder is ~/Work/Projects"

Later: "Create a new project"
AI: [Uses ~/Work/Projects automatically]

What to store:

• Preferred tools/languages
• Common paths
• Work patterns
• Preferences

See Memory System guide.

Use Multi-Step Plans

For complex workflows:

"Organize my Downloads:
1. Find files older than 60 days
2. Group by type
3. Show me the groups
4. Move to Archives after I confirm
5. Generate a summary report"

Benefits:

• Visual progress
• Control at each step
• Easy to pause/modify

See Plans Guide.

Create Reusable Automations

For recurring tasks:

First time: "Write a script to backup my Documents folder"
           [Test and verify]
           "Save this as 'Daily Backup' automation"

Every day: "Run daily backup"

See Automations.

Attach Images for Visual Context

Don't describe, show:

❌ "There's an error message that says something about modules not found..."

✅ [Attach screenshot of error]
    "How do I fix this?"

See Attachments Guide.

Conversation Best Practices

Start Fresh for New Topics

New conversation when:

• Switching to unrelated topic
• Previous conversation got long (>50 messages)
• Need clean context
• Starting complex project

Continue conversation when:

• Follow-up questions
• Related tasks
• Building on previous work
• Iterating on solution

Give Feedback

Help AI improve:

✅ "That's perfect, thank you"
✅ "This isn't quite right - I meant..."
✅ "Close, but can you adjust to..."
✅ "This worked great, do the same for..."

AI learns from feedback and adjusts approach

Ask for Explanations

Understand before executing:

"Explain what this command will do before running it"
"Why did you choose this approach?"
"What are the risks of this operation?"
"Can you break down what happened?"

Use Follow-Ups Effectively

Reference previous context:

AI: [Lists 50 files]

You: "Show only the largest 10"
AI: [Filters]

You: "Delete those"
AI: [Prepares deletion of the 10 files]

AI remembers what "those" refers to

Choosing the Right Model

Match Model to Task

Flash Lite:

• Quick questions
• Simple file listings
• Basic searches
• When speed matters most

Flash (default):

• Most everyday tasks
• File management
• General automation
• Best balance

Pro:

• Complex analysis
• Large codebases
• Detailed planning
• When quality matters most

Devstral 2 (Pro):

• Heavy code work
• Technical automation
• Development workflows

See Model Selection guide.

Switch Mid-Task

If results unsatisfactory:

"Switch to Gemini Pro and retry that request"
"Use a more powerful model for this task"

Safety & Control

Use Safe Mode While Learning

Benefits:

• See what AI will do before it happens
• Learn safe vs risky operations
• Build trust
• Catch mistakes

Switch to Full Mode when:

• You understand AI's behavior
• Running trusted automations
• Speed is important

See Safety Guide.

Review Confirmations Carefully

Don't just click "Proceed":

⚠️ About to delete 47 files

Read:
• Which files?
• Are these correct?
• Is this what you wanted?

Then decide: Proceed or Cancel

Request Backups First

For important operations:

✅ "Before organizing these files, create a backup"
✅ "Copy these to a safe location first"

Performance Optimization

Keep Conversations Focused

Long conversations slow down:

• More context = more tokens
• Slower responses
• Higher costs
• Potential errors

Solution:

• Start new chat every 50-100 messages
• Or when switching topics

Use Specific Paths

Instead of:

❌ "Find my project"

Use:

✅ "List files in ~/Work/my-project"

Faster because:

• No ambiguity
• No searching
• Direct action

Batch Related Requests

Instead of multiple messages:

❌ "Find PDFs"
   "Now find images"
   "Now find videos"

Batch:

✅ "Find all PDFs, images, and videos in my Downloads folder"

Troubleshooting Techniques

When AI Misunderstands

Try:

1. Rephrase more explicitly
2. Provide examples
3. Break into smaller steps
4. Start new conversation (reset context)

When Results Are Wrong

Debug:

"Why did you choose this approach?"
"What made you think X?"
"Show me your reasoning"

Then correct:

"Actually, I meant Y not X"
"Let's try a different approach"

When Operations Fail

Gather info:

"What went wrong?"
"Show me the error details"
"Why did this fail?"

AI can often:

• Explain the error
• Suggest fixes
• Try alternative approach

Advanced Techniques

Chain Operations

Build workflows:

"Find large files, then show me the top 10, then help me decide what to delete"

AI executes sequentially

Conditional Logic

Ask AI to decide:

"If disk space is >80% full, find and suggest files to delete.
 Otherwise, just report current usage."

Comparative Analysis

Use multiple files:

[Attach: config-old.json, config-new.json]
"What changed between these configs?"

Template Generation

Create reusable patterns:

"Generate a README template for my Python projects"
[Save as automation]

Later: "Use my Python README template for this project"

Learning from Results

Study AI Responses

When AI does something well:

"Explain how you did that"
"Can we apply this approach to X?"
"Save this as an automation for future use"

Analyze Failures

When something fails:

"What would have been a better way to phrase that request?"
"How should I ask next time?"

AI can teach you to communicate better

Common Mistakes to Avoid

❌ Being Too Vague

Bad: "Fix my computer"
Good: "My laptop fans are running loud. Show me which processes are using high CPU."

❌ Ignoring Errors

Bad: [Error appears] → Try same thing again
Good: "What caused this error? How can I fix it?"

❌ Not Providing Context

Bad: "This doesn't work" [no other info]
Good: "This Python script throws ImportError. I'm using Python 3.9 and pip installed the package."

❌ Overusing Full Mode

Bad: Full Mode for everything → Accidents happen
Good: Full Mode for trusted workflows, Safe Mode for new/risky tasks

❌ Not Using Memory

Bad: Repeating same preferences every chat
Good: Store preferences once, AI remembers

❌ Mega-Conversations

Bad: One conversation with 500 messages spanning weeks
Good: New conversation for new topics/projects

Quick Reference Card

For best results:

✅ Be specific - Include details, criteria, locations
✅ Provide context - Explain what you're trying to do
✅ Use natural language - Speak normally, not in commands
✅ Iterate - Start broad, refine as needed
✅ Ask for previews - Review before executing
✅ Use memory - Store frequently used info
✅ Create automations - Reuse successful workflows
✅ Choose right model - Match model to task complexity
✅ Stay safe - Use Safe Mode while learning
✅ Give feedback - Help AI understand what you want

❌ Avoid vague requests - "Fix stuff"
❌ Skip context - Just error text, no background
❌ Ignore confirmations - Read before clicking Proceed
❌ Forget to review - Check results before trusting
❌ Make mega-conversations - Start fresh for new topics

Next: Read about Prompt Engineering for advanced techniques.