AIWB Workflow Guide - From Beginner to Power User

🎯 Purpose

This guide shows you different workflows for using AIWB, from simple one-liners to advanced multi-model refinement loops. Learn by example!

📚 Table of Contents

Quick Start - Your First Generation
The Old Way vs The New Way
Example Workflows
Understanding the Models
Cost Management
Advanced Techniques
Free APIs Worth Using

🚀 Quick Start - Your First Generation

The Absolute Easiest Way (Recommended!)

# One command, that's it!
./aiwb quick "Create a password generator script"

What happens automatically:

✅ Creates your script
✅ Verifies it with a different AI model (catches errors!)
✅ Shows you both the code and the review
✅ Saves everything to outputs
✅ Cleans up temporary files

Total typing: One command. Done!

Alternative: Interactive Wizard

./aiwb wizard

# Then just answer a few questions:
# - What do you want to build? → Simple script/tool
# - Brief description → Password generator
# - Technology? → Bash/Shell
# - Special requirements? → Exclude confusing characters
# - Proceed? → Yes
# - Workflow? → Quick (generate + verify)

Great for: When you want a bit more structure and guidance.

🔄 Old Way vs New Way

❌ THE OLD TEDIOUS WAY (Before our improvements)

# Step 1: Create a project (do I really need this?)
./aiwb project create password-tool

# Step 2: Switch to project (ugh)
./aiwb project switch password-tool

# Step 3: Create a task (more setup...)
./aiwb task create build-password-gen
# Opens editor, write prompt, save, exit

# Step 4: Estimate cost (okay this is useful)
./aiwb cpre build-password-gen

# Step 5: Generate (finally!)
./aiwb cgo build-password-gen

# Step 6: Manually verify (even more typing)
./aiwb task create verify-password-gen
# Opens editor AGAIN, write review prompt, save
./aiwb ggo verify-password-gen

# Step 7: Check outputs (where is everything?)
./aiwb outputs list

# Step 8: View the file
cat ~/.aiwb/workspace/projects/password-tool/genpass.sh

Total steps: 8+ commands, multiple manual edits Time: 5-10 minutes Cognitive load: HIGH - need to remember task names, file paths, etc.

✅ THE NEW STREAMLINED WAY

# One command
./aiwb quick "Create a password generator script"

# Optional: Improve it
./aiwb improve "Exclude ambiguous characters like 0,O,l,1"

# Optional: Improve again
./aiwb improve "Add a password strength meter"

Total steps: 1 command (+ optional improvements) Time: 30 seconds Cognitive load: MINIMAL - just describe what you want

💡 Example Workflows

Example 1: Building a Simple Tool

Goal: Create a JSON formatter script

Quick Method (30 seconds)

./aiwb quick "Create a bash script that formats JSON from stdin or file, with color syntax highlighting"

Output:

~/.aiwb/workspace/outputs/quick_1234567890.md - Generated script
~/.aiwb/workspace/outputs/quick_1234567890.feedback.md - AI review

With Improvements (2 minutes)

# Initial version
./aiwb quick "JSON formatter with color highlighting"

# Add features iteratively
./aiwb improve "Add option to minify JSON instead of prettify"
./aiwb improve "Add validation with helpful error messages"
./aiwb improve "Support reading from URLs via curl"

# Check what we spent
./aiwb costs

Example 2: Learning a New Technology

Goal: Understand how to use the GitHub API

Interactive Approach

./aiwb

# In the chat session:
aiwb> Explain how to authenticate with GitHub API and show a simple example in bash
aiwb> Now show me how to list all my repositories
aiwb> How do I create a new issue programmatically?
aiwb> /make "Complete GitHub API client in bash with auth, list repos, create issues"
aiwb> /improve "Add error handling and rate limit checks"
aiwb> /costs

Learning flow:

Ask questions to understand concepts
Generate working code when ready
Improve iteratively
Track costs

Example 3: Rapid Prototyping

Goal: Try 3 different approaches to the same problem

# Approach 1: Pure bash
./aiwb quick "Create a file backup script in pure bash"

# Approach 2: Python
./aiwb quick "Create a file backup script in Python with argparse"

# Approach 3: Using rsync
./aiwb quick "Create a file backup script wrapping rsync with smart defaults"

# Now compare the outputs
ls -lh ~/.aiwb/workspace/outputs/

# Pick the best and improve it
./aiwb improve "Add incremental backup support and logging"

Example 4: Using Templates for Best Practices

# Browse available templates
./aiwb templates

# Select "bash-script" template
# It pre-fills best practices:
# - POSIX compliance
# - Error handling
# - Input validation
# - Help messages
# - Colorful output

# Just add your specific requirements
# Template asks: "Describe what the script should do"
# You answer: "Monitor system resources and alert if threshold exceeded"

# Result: Professional-grade script with all best practices baked in

🤖 Understanding the Models

Which AI Should I Use?

AIworkbench supports multiple providers. Here's the quick guide:

For Quick Scripts & Tools

Gemini 2.5 Flash (Default) - Fast, cheap, great for code
Groq Llama 3.3 70B - FREE, blazing fast, good quality

For Complex Logic

Claude 3.5 Sonnet - Best reasoning, excellent code quality
GPT-4o - Strong all-rounder

For Verification (Cross-Check)

AIWB automatically uses a different model to verify. Example:

You generate with Gemini → Verifies with Claude
You generate with Claude → Verifies with Gemini

Why? Different AIs catch different mistakes!

Switching Providers

# Interactive mode
./aiwb
aiwb> /models
# Choose Provider → groq
# Choose Model → llama-3.3-70b-versatile

# Or direct
./aiwb --provider groq --model llama-3.3-70b-versatile quick "Create a script"

💰 Cost Management

See What You're Spending

# Quick cost check
./aiwb costs

# Output shows:
# Total spent: $0.05 USD
#
# By Provider:
# gemini    $0.03
# claude    $0.02
#
# Recent activity:
# 2025-11-09    gemini    $0.01
# 2025-11-09    claude    $0.02

Smart Cost Estimation

Before running expensive operations:

# Estimate a task
./aiwb estimate my-big-task

# Shows:
# Tier      Output Tokens    Cost (USD)
# ───────────────────────────────────────
# Basic     1300             $0.02
# Medium    2000             $0.03
# Best      3200             $0.05

Auto-Estimate Settings

./aiwb
aiwb> /models
# → Preferences
# → Auto-estimate: true (estimates before generating)
# → Confirm before generate: true (asks if cost > threshold)
# → Show costs: true (shows cost after each operation)

Using Free APIs to Save Money

Switch to Groq (100% free!) for development:

./aiwb settings
# Provider → groq
# Model → llama-3.3-70b-versatile

# Or keep Gemini (also has generous free tier)

🎓 Advanced Techniques

Multi-Iteration Refinement

For complex projects, use the refinement loop:

./aiwb
aiwb> /refine

# This automatically:
# 1. Generates code
# 2. Verifies with different AI
# 3. Incorporates feedback
# 4. Generates improved version
# 5. Verifies again
# 6. Repeat 3 times (or custom iterations)

Conversation History

# View past sessions
./aiwb history

# Use fzf to search and view any session
# Or see last 10:
ls -lht ~/.aiwb/workspace/logs/chat_*.log | head -10

Template Customization

Create your own templates:

# 1. Create template file
cat > ~/.aiwb/workspace/templates/my-api-template.md <<'EOF'
Create a REST API with:
- Express.js framework
- JWT authentication
- MongoDB database
- Input validation with Joi
- Error handling middleware
- API documentation with Swagger

Describe the API endpoints:
EOF

# 2. Use it
./aiwb templates
# Select: my-api-template

Combining Direct and Interactive

# Generate something quickly
./aiwb quick "Create a log analyzer script"

# Then enter interactive mode to refine
./aiwb
aiwb> /improve "Add support for multiple log formats"
aiwb> /improve "Add statistics and charts using gnuplot"
aiwb> How do I deploy this to a server?
aiwb> /make "Create a systemd service file for the log analyzer"

🌟 Free APIs Worth Using

Already Supported & Free

Groq ⚡
- Status: 100% FREE
- Tokens/min: 6,000 TPM
- Requests/day: 14,400
- Speed: Blazing fast (LPU architecture)
- Models: Llama 3.3 70B, Mixtral 8x7B, Gemma 2 9B
- Already in AIWB!
Google Gemini (Free Tier)
- Status: Generous free tier
- Models: 2.5 Flash, 2.0 Flash Lite (very fast & cheap)
- Great for: Code generation, quick tasks
- Already in AIWB! (Default)
Anthropic Claude (Free Trial)
- Status: $5 free credits for new accounts
- Models: Claude 3.5 Sonnet, 3.5 Haiku
- Best for: Complex reasoning, code quality
- Already in AIWB!

Recommended to Add

Hugging Face Inference API ⭐⭐⭐⭐⭐
- Status: FREE with limits
- Models: 100,000+ open-source models
- Why add: Massive variety, true OSS
- Use case: Experimentation, specialized models
Mistral AI
- Status: FREE tier available
- Models: Mistral 7B, Mixtral 8x7B
- Why add: European option, great performance
- Use case: Alternative to big US providers
Cohere
- Status: FREE for prototyping
- Models: Command R, Command R+
- Why add: Best-in-class for RAG
- Use case: Retrieval-augmented tasks

Cost Comparison (Paid Tiers)

For when you outgrow free tiers:

Provider	Model	Input ($/1M tokens)	Output ($/1M tokens)
Groq	Llama 3.3 70B	$0.59	$0.79
Gemini	2.5 Flash	$0.075	$0.30
Claude	3.5 Sonnet	$3.00	$15.00
OpenAI	GPT-4o	$2.50	$10.00

Tip: Use Groq/Gemini for iterations, Claude for final polish!

🎯 Recommended Workflow Progression

Week 1: Learn the Basics

# Day 1: Setup
./aiwb keys  # Add your API keys
./aiwb doctor  # Check everything works

# Day 2-3: Try quick commands
./aiwb quick "Create a hello world script"
./aiwb quick "Create a file renamer tool"
./aiwb quick "Create a API request script"

# Day 4-5: Use improvements
./aiwb quick "Create a backup script"
./aiwb improve "Add logging"
./aiwb improve "Add email notifications"

# Day 6-7: Explore templates
./aiwb templates

Week 2: Get Comfortable

# Use wizard for structured projects
./aiwb wizard

# Try different providers
./aiwb settings  # Switch to Claude
./aiwb quick "Complex algorithm implementation"

./aiwb settings  # Switch to Groq
./aiwb quick "Quick utility script"

# Practice cost awareness
./aiwb costs  # Check spending

Week 3: Advanced Usage

# Multi-iteration refinement
./aiwb refine

# Custom templates
# Create templates for your common patterns

# Integrate into your workflow
# Use AIWB for all scripting needs

Month 2+: Power User

# Combine everything:
# - Use templates for structure
# - Quick for rapid prototyping
# - Improve for iterations
# - Different models for different tasks
# - Cost-optimize with free APIs

# Example daily usage:
./aiwb quick "Today's automation task"
./aiwb improve "Based on testing feedback"
./aiwb costs  # Stay on budget

📖 Quick Reference Card

# ONE-LINERS (Use These Daily!)
./aiwb quick "description"     # Generate + verify
./aiwb improve "suggestion"    # Improve last output
./aiwb costs                   # Check spending

# INTERACTIVE
./aiwb                         # Start chat session
  /quick description           # Quick generate
  /wizard                      # Guided workflow
  /improve suggestion          # Improve last
  /templates                   # Browse templates
  /models                      # Change provider/model
  /costs                       # Cost check
  /help                        # Show all commands

# CONFIGURATION
./aiwb keys                    # Manage API keys
./aiwb settings                # Configure everything
./aiwb status                  # Current setup

# ADVANCED
./aiwb refine                  # Multi-iteration loop
./aiwb wizard                  # Step-by-step guide
./aiwb templates               # Use/create templates

🤝 Getting Help

Help command: ./aiwb --help or /help in chat
Check health: ./aiwb doctor
Current status: ./aiwb status
GitHub: https://github.com/juanitto-maker/AIworkbench

🎓 Learning Resources

Understand the Philosophy

AIWB is built on the Generator-Verifier Loop concept:

One AI generates code
A different AI verifies it
Feedback improves the next iteration
Repeat until satisfied

This catches more bugs than single-model generation!

Best Practices

Start small: Use /quick for everything at first
Iterate freely: Use /improve without fear - it's cheap with free APIs
Cross-check: Let auto-verify catch mistakes
Track costs: Run /costs weekly to stay aware
Use templates: They encode best practices
Switch models: Different AIs for different tasks

Common Patterns

# Pattern 1: Rapid prototype
./aiwb quick "idea" → test → improve → improve → done

# Pattern 2: Exploration
./aiwb → chat to understand → /quick when ready → iterate

# Pattern 3: Production-quality
./aiwb wizard → choose "Full refinement" → review outputs → done

# Pattern 4: Learning
./aiwb → ask questions → generate examples → improve understanding

Remember: The goal is LESS typing, MORE automation, BETTER results.

Start with /quick and add complexity only when you need it! 🚀

Uh oh!

FilesExpand file tree

WORKFLOW-GUIDE.md

Latest commit

History