filoma 1.12.4

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) ⚡🍃

Whether you're auditing a machine-learning dataset, tracking down duplicates across terabytes, or just need a quick overview of what's in a directory — Filoma gives you the tools to go from raw folder structure to actionable insight in seconds.

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 →

� Talk to your filesystem: filoma filaraki chat — ask questions about your data in plain English. Find duplicates, audit datasets, export HTML reports — all from one conversation. Try it →

�🎯 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 with filesystem-specific operations for filtering, grouping, and summarization.

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

# Filter and analyze
df.filter_by_extension([".py", ".rs"])
df.extension_counts()
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: parent, name, stem, suffix, size_bytes, modified_time,
   created_time, is_file, is_dir, owner, group, mode_str, inode, nlink, sha256, xattrs, depth

🔍 See Operation Examples

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   │
└────────────┴─────┘

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

---

🗂️ 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 ("little leaf" / "little buddy" in Greek) is Filoma's agentic interface for natural language filesystem analysis. Available as an interactive chat CLI, programmatic API, or MCP server.

Interactive Chat CLI

filoma filaraki chat

Programmatic Usage

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("Sort dataframe by size descending and show top 5")

AI Service Options

Filaraki supports multiple providers — pick whatever fits your setup:

Provider	Requires	Privacy
Ollama (default)	ollama serve on localhost:11434	🔒 100% local
Mistral AI	MISTRAL_API_KEY	Cloud
Google Gemini	GEMINI_API_KEY	Cloud
OpenAI / OpenRouter / compatible	FILOMA_FILARAKI_BASE_URL + OPENAI_API_KEY	Cloud

📖 Full AI configuration guide →

🏠 Local AI Setup (Nanobot + Ollama)

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

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

This installs nanobot + Ollama with Filoma's 21 filesystem tools. No API keys, no cloud — everything stays on your machine.

📖 Full MCP Configuration Guide →

📊 One-Command Audit with HTML Report

Run a full audit and export a self-contained interactive HTML report in one prompt:

filoma filaraki chat
> perform an audit on /path/to/dataset and export an html report called audit.html

📝 What's in the report?

• Score gauges for Hygiene and Migration Readiness
• KPI strip showing file counts, duplicate groups, and space waste
• Stage timing bars (integrity / hygiene / readiness)
• Priority-tagged Next Actions — colour-coded high / medium / low
• Duplicate evidence cards with exact file paths
• Collapsible full JSON payload for deeper inspection

Export formats: html, json, md

MCP Server

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

filoma mcp serve

📖 Browse all guides →

---

📊 Performance & Benchmarks

Backend	Local SSD (1M files)	Network (200K files)
🦀 Rust	7.3s — 136K files/sec	2.3s — 86K files/sec
⚡ Async	11.5s — 87K files/sec	2.8s — 70K files/sec
🐍 Python	35.5s — 28K files/sec	15.1s — 13K files/sec

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

📖 Full Benchmarks Guide →

---

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.