filoma 1.12.2

Modular Python tool for profiling files, analyzing directory structures, and inspecting image data

Fast, multi-backend file/directory profiling and data preparation.

pip install filoma

Installation • Documentation • Agentic Analysis • Interactive CLI • Quickstart • Cookbook • Roboflow Demo • Source Code

📖 New to Filoma? Check out the Cookbook for practical, copy-paste recipes for common tasks!

---

filoma helps you analyze file directory trees, inspect file metadata, and prepare your data for exploration. It can achieve this blazingly fast using the best available backend (Rust, fd, or pure Python) ⚡🍃

Key Features

• 🚀 High-Performance Backends: Automatic selection of Rust, fd, or Python for the best performance.
• 📈 DataFrame Integration: Convert scan results to Polars (or pandas) DataFrames for powerful analysis.
• 📊 Rich Directory Analysis: Get detailed statistics on file counts, extensions, sizes, and more.
• 🔍 Smart File Search: Use regex and glob patterns to find files with FdFinder.
• 🖼️ File/Image Profiling: Extract metadata and statistics from various file formats.
• 🛡️ Dataset Integrity & Quality: Unified integrity checking for snapshots, manifests, and automated quality scans (corruption, duplicates, leakage, class balance). 📖 Data Integrity Guide →
• 🧠 Agentic Analysis: Natural language interface for file discovery, deduplication, and metadata inspection. 📖 Filaraki Guide →
• 🖥️ Interactive CLI: Beautiful terminal interface for filesystem exploration and DataFrame analysis. 📖 CLI Documentation →
• 🌐 MCP Server: Expose all 21 filesystem tools to any MCP-compatible AI assistant (nanobot recommended). 📖 MCP Configuration →

🎯 Local AI in 10 seconds: curl -sL https://raw.githubusercontent.com/kalfasyan/filoma/main/scripts/install.sh | sh → Use with nanobot + Ollama for fully local filesystem analysis. Learn more →

---

⚡ Quick Start

filoma provides a unified API for filesystem analysis.

End-to-End Example: Folder → DataFrame → Insights

This is the core Filoma workflow in one place: scan a folder, build a rich dataframe, filter it, and extract quick insights.

import filoma as flm

dataset = "notebooks/Weeds-3"

# 1) Fast scan + high-level summary
analysis = flm.probe(dataset)
analysis.print_summary()

# 2) Build an enriched dataframe (paths, extension, sizes, ownership, timestamps, etc.)
df = flm.probe_to_df(dataset, enrich=True)

# 3) Narrow to image files and inspect distribution
images = df.filter_by_extension(["jpg", "png"])
print(images.extension_counts())
print(images.directory_counts().head(3))

# 4) Get the largest files quickly
largest = images.sort("size_bytes", descending=True).head(5)
print(largest.select(["path", "size_bytes"]))

This flow is typically the fastest way to move from raw folder structure to actionable dataset insight.

1. File & Image Profiling

Extract rich metadata and statistics from any file or image.

import filoma as flm

# Profile any file
info = flm.probe_file("README.md")
print(info)

📄 See Metadata Output

Filo(
    path=PosixPath('README.md'),
    size=12237,
    mode_str='-rw-rw-r--',
    owner='user',
    modified=datetime.datetime(2025, 12, 30, 22, 45, 53),
    is_file=True,
    ...
)

For images, probe_image automatically extracts shapes, types, and pixel statistics.

2. Directory Analysis

Scan entire directory trees in milliseconds. filoma automatically picks the fastest available backend (Rust → fd → Python).

# Analyze a directory
analysis = flm.probe('.')

# Print high-level summary
analysis.print_summary()

📂 See Directory Summary Table

 Directory Analysis: /project (🦀 Rust (Parallel)) - 0.60s
┏━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━┓
┃ Metric                   ┃ Value                ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━┩
│ Total Files              │ 57,225               │
│ Total Folders            │ 3,427                │
│ Total Size               │ 2,084.90 MB          │
│ Average Files per Folder │ 16.70                │
│ Maximum Depth            │ 14                   │
│ Empty Folders            │ 103                  │
│ Analysis Time            │ 0.60s                │
│ Processing Speed         │ 102,114 items/sec    │
└──────────────────────────┴──────────────────────┘

# Or get a detailed report with extensions and folder stats
analysis.print_report()

📊 See Detailed Directory Report

          File Extensions
┏━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━┓
┃ Extension  ┃ Count  ┃ Percentage ┃
┡━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━┩
│ .py        │ 240    │ 12.8%      │
│ .jpg       │ 1,204  │ 64.2%      │
│ .json      │ 431    │ 23.0%      │
│ .svg       │ 28,674 │ 50.1%      │
└────────────┴────────┴────────────┘

          Common Folder Names
┏━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┓
┃ Folder Name   ┃ Occurrences ┃
┡━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━┩
│ src           │ 1           │
│ tests         │ 1           │
│ docs          │ 1           │
│ notebooks     │ 1           │
└───────────────┴─────────────┘

          Empty Folders (3 found)
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Path                                       ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ /project/data/raw/empty_set_A              │
│ /project/logs/old/unused                   │
│ /project/temp/scratch                      │
└────────────────────────────────────────────┘

3. DataFrame Analysis

Convert scan results to Polars DataFrames for advanced analysis.

# Scan and get an enriched filoma.DataFrame (Polars)
df = flm.probe_to_df('src', enrich=True)

# Perform operations
df.filter_by_extension([".py", ".rs"])
df.directory_counts()

📊 See Enriched DataFrame Output

filoma.DataFrame with 2 rows
shape: (2, 18)
┌───────────────────┬───────┬────────┬───────────────┬───┬─────────┬───────┬────────┬────────┐
│ path              ┆ depth ┆ parent ┆ name          ┆ … ┆ inode   ┆ nlink ┆ sha256 ┆ xattrs │
│ ---               ┆ ---   ┆ ---    ┆ ---           ┆   ┆ ---     ┆ ---   ┆ ---    ┆ ---    │
│ str               ┆ i64   ┆ str    ┆ str           ┆   ┆ i64     ┆ i64   ┆ str    ┆ str    │
╞═══════════════════╪═══════╪════════╪═══════════════╪═══╪═════════╪═══════╪════════╪════════╡
│ src/async_scan.rs ┆ 1     ┆ src    ┆ async_scan.rs ┆ … ┆ 7601121 ┆ 1     ┆ null   ┆ {}     │
│ src/filoma        ┆ 1     ┆ src    ┆ filoma        ┆ … ┆ 7603126 ┆ 8     ┆ null   ┆ {}     │
└───────────────────┴───────┴────────┴───────────────┴───┴─────────┴───────┴────────┴────────┘

✨ Enriched columns added: parent, name, stem, suffix, size_bytes, modified_time,
   created_time, is_file, is_dir, owner, group, mode_str, inode, nlink, sha256, xattrs, depth

• Seamless Pandas Integration: Just use df.pandas for instant conversion.
• Lazy Loading: import filoma is cheap; heavy dependencies load only when needed.

4. Specialized DataFrame Operations

Filoma's DataFrame extends Polars with filesystem-specific operations for quick filtering and summarization.

# Filter by extensions
df.filter_by_extension([".py", ".rs"])

# Quick frequency analysis
df.extension_counts()
df.directory_counts()

🔍 See Operation Examples

filter_by_extension([".py", ".rs"])

shape: (3, 1)
┌─────────────────────┐
│ path                │
│ ---                 │
│ str                 │
╞═════════════════════╡
│ src/async_scan.rs   │
│ src/lib.rs          │
│ src/filoma/dedup.py │
└─────────────────────┘

extension_counts() — groups files by extension and returns counts.

shape: (3, 2)
┌────────────┬─────┐
│ extension  ┆ len │
│ ---        ┆ --- │
│ str        ┆ u32 │
╞════════════╪═════╡
│ .py        ┆ 240 │
│ .jpg       ┆ 124 │
│ .json      ┆ 43  │
└────────────┴─────┘

directory_counts() — summarizes file distribution across parent directories.

shape: (3, 2)
┌────────────┬─────┐
│ parent_dir ┆ len │
│ ---        ┆ --- │
│ str        ┆ u32 │
╞════════════╪═════╡
│ src/filoma ┆ 12  │
│ tests      ┆ 8   │
│ docs       ┆ 5   │
└────────────┴─────┘

---

🗂️ Advanced Topics

Dataset Convenience Class

Use the Dataset class for orchestration of snapshotting, profiling, integrity checks, and AI interactions:

import filoma as flm

ds = flm.Dataset("./my_data")

# Snapshot, Quality Scan, and Deduplication
ds.snap(mode="deep")
ds.run_quality_scan()
ds.dedup()

# Get an enriched DataFrame of the dataset
df = ds.to_dataframe()
print(df.extension_counts())

# Agentic interaction with this specific dataset
ds.get_filaraki().run("Is there any class imbalance in my dataset?")

Dataset Integrity & Quality

Filoma provides a comprehensive suite for dataset validation (corruption, leaks, balance) and manifest integrity:

from filoma.core.verifier import DatasetVerifier
verifier = DatasetVerifier("./data")
verifier.run_all()
verifier.print_summary()

Deduplication

Find duplicate files, images (perceptual hash), or text files.

# Standard find
filoma dedup /path/to/dataset

# Cross-directory find
filoma dedup train/ valid/ --cross-dir

🍃 Agentic Analysis

Filaraki (stands for "little leaf" or "little buddy" in Greek) is Filoma's agentic interface for natural language filesystem analysis. It provides an agentic, flexible way to interact with your data using plain language commands.

Connect your "filaraki" agent to your filesystem for natural language interaction. Available as an interactive chat CLI, programmatic API, or MCP server for integration with AI assistants.

🏠 Local AI Setup (Nanobot + Ollama)

Run Filoma Filaraki completely offline with local models via the MCP server:

# One-command setup
curl -sL https://raw.githubusercontent.com/kalfasyan/filoma/main/scripts/install.sh | sh

This installs nanobot and configures it to use Ollama with Filoma's 21 filesystem tools. No API keys, no cloud services—everything stays on your machine.

# After setup, chat with your filesystem
nanobot agent -m "How many images are in ./data?"
nanobot agent -m "Find duplicate files and show me the largest ones"

📖 Full MCP Configuration Guide →

Interactive Chat CLI

Start a chat session directly from your terminal:

filoma filaraki chat

Programmatic Usage

Use Python for scripted workflows:

from filoma.filaraki import get_agent

agent = get_agent()
await agent.run("Create a dataframe from notebooks/Weeds-3 with enrichment")
await agent.run("Filter by extension: jpg, png")
await agent.run("Summarize dataframe and show top directories")
await agent.run("Sort dataframe by size descending and show top 5")

Advanced Workflow Orchestration

Filoma Filaraki includes advanced orchestrator tools for enterprise-grade dataset analysis:

# Run advanced workflow examples
make filaraki-advanced

# Or in code:
await agent.run("Run a corrupted file audit on /path/to/dataset")
await agent.run("Generate a dataset hygiene report for /path/to/dataset")
await agent.run("Assess the migration readiness of /path/to/dataset")

These provide structured, deterministic reports with detailed findings, recommendations, and confidence scores.

MCP Server

Expose all 21 filesystem tools to any MCP-compatible client:

filoma mcp serve

📖 Browse all guides →

---

📊 Performance & Benchmarks

Need to compare backend performance? Check out the comprehensive Benchmarks Guide!

Local SSD (1M files):

• 🦀 Rust: 7.3s (136K files/sec)
• ⚡ Async: 11.5s (87K files/sec)
• 🐍 Python: 35.5s (28K files/sec)

Network Storage (200K files, cold cache):

• 🦀 Rust: 2.3s (86K files/sec)
• ⚡ Async: 2.8s (70K files/sec)
• 🐍 Python: 15.1s (13K files/sec)

python benchmarks/benchmark.py --path /your/directory -n 3 --backend profiling

---

License

This work is licensed under a Creative Commons Attribution 4.0 International License.

---

Contributing

Contributions welcome! Please check the issues for planned features and bug reports.