Project Summarizer Agent

AI-powered tool to automatically assess and summarize multiple coding projects using Ollama-hosted models.

What's New

Fixed Issues

• Schema alignment - Template now matches markdown output (added status, complexity, tags fields)
• Config loading - Now properly loads from config.toml instead of hardcoded values
• Field type fix - Changed percentage_complete from array to integer (0-100)

Enhancements

1. TOML Configuration - Externalized all settings to config.toml
2. Directory Exclusions - Automatically skips .git, node_modules, venv, etc.
3. Retry Logic - Exponential backoff (1s, 2s, 4s) for failed API calls
4. Incremental Processing - Skips projects processed within 7 days
5. Context Warnings - Alerts when project files exceed safe token limits
6. Better Logging - Structured logging with timestamps and levels
7. Progress Tracking - Shows "Processing 5/23 projects..."
8. JSON Validation - Verifies model responses match expected schema
9. Summary Statistics - Average completion, status/complexity distributions
10. Concurrent Processing - Processes 3 projects in parallel (configurable)

Installation

# For Python 3.11+
pip install requests

# For Python <3.11, also install:
pip install tomli

Security

Important: If you're using an API key for cloud-hosted models:

1. Never commit config.toml to version control if it contains your API key
2. Add config.toml to your .gitignore
3. Consider using environment variables instead:
   • Set OLLAMA_API_KEY environment variable
   • Remove api_key from config.toml
   • The code will check environment variables as a fallback

Configuration

Edit config.toml:

[project]
root = "D:/Projects"              # Legacy single-root option
roots = ["D:/Projects"]           # Preferred: evaluate each top-level child directory under these roots
expand_children_of = ["D:/Projects/utilities"]  # Treat each child folder here as its own project
standalone_projects = [           # Explicit project roots outside the main roots
  "C:/Users/You/Desktop/Tool & Data Tracking",
  "C:/Users/You/Desktop/Aptlantis-Studio",
]
out_dir = "D:/Projects/project_summaries"

[model]
name = "minimax-m2:cloud"         # Ollama model name
ollama_host = "http://127.0.0.1:11434"
ollama_api_key = "your-api-key-here"  # Optional: for cloud-hosted models

[processing]
max_chars_per_file = 2500         # Truncate long files
max_total_chars = 90000           # Hard cap for total prompt context per project
max_files_per_project = 60        # Max number of files sampled per project
delay = 0.3                       # Delay between API calls (seconds)
max_retries = 3                   # Retry failed requests
skip_already_processed = true     # Skip recently processed projects
max_workers = 3                   # Concurrent processing threads

[filtering]
allowed_exts = [".py", ".md", ...]  # File types to analyze
excluded_dirs = [".git", "node_modules", ...]  # Directories to skip

Usage

python Summarizer.py

Output:

• Individual JSON files: project_summaries/{project_name}.json
• Combined markdown: project_summaries/summary.md

Schema

Each project is assessed with:

{
  "project": "MyProject",
  "source_path": "D:/Projects/MyProject",
  "summary": "Brief description...",
  "status": "In Progress",
  "percentage_complete": 75,
  "missing_pieces": ["Unit tests", "Documentation"],
  "next_steps": ["Add error handling", "Deploy to prod"],
  "potential_improvements": ["Add caching", "Optimize queries"],
  "complexity": "Medium",
  "tags": ["python", "web", "automation"]
}

Features

• Smart Filtering - Skips binary files, build artifacts, dependencies
• Budgeted Context - Caps total chars and sampled files so very large repos do not blow up the model request
• Flexible Discovery - Supports multiple roots, explicit standalone projects, and expanded child-project folders
• Incremental Updates - Only reprocesses changed/new projects
• Fault Tolerant - Continues processing even if individual projects fail
• Rich Output - Markdown summary sorted by completion percentage
• Statistics Dashboard - Overview of all projects at a glance

Performance

• Processes 20 small projects in ~2-3 minutes (with 3 workers)
• Uses ~100-500 tokens per project depending on size
• Automatic caching reduces re-analysis of unchanged projects