Mongotar
Serializes a whole directory into one human-readable text file for AI prompts, then rebuilds it back to disk
🔗 Visit Mongotar
Description
When you want an AI model to understand your whole project, you either paste files one by one or zip them up as binary — neither of which the model can really read cleanly. Mongotar turns a folder into a single plain-text archive the model can read like a book, then turns that text back into real files afterward.
Mongotar serializes an entire directory tree into one human-readable text archive, preserving file paths and basic permissions (rwx flags) and respecting .gitignore rules. It's aimed squarely at LLM workflows, where a developer needs to bundle code or data into a prompt-friendly format. It works both ways: serialize to a text archive via CLI or Python library, then deserialize back to disk with safe defaults that refuse to overwrite existing files unless told to. It supports glob-based inclusion and exclusion rules for selecting exactly what goes in.
💬 Our review
The short version: it does one small job — files to text and back — cleanly, but it's a two-star, single-contributor project competing against more established tools doing the same thing.
Compared to files-to-prompt or repomix, which solve the same "bundle my repo for an LLM" problem, Mongotar's differentiator is the round-trip: it can deserialize its own archive back into real files, not just produce one-way output. The tradeoffs are real limitations: it only handles UTF-8 text (no binaries or symlinks), its permission model is simplified to user-only bits, and it breaks on files that happen to contain its own header-format strings. Free and worth trying if you specifically need the round-trip capability; if you only need one-way bundling, the more established alternatives have wider adoption.
📊 Global score
🤖 AI-enriched data
Gratuit, open source. CLI et librairie Python.
Pros
Format texte lisible par un humain, idéal pour l'injecter dans un prompt LLM
Sérialisation bidirectionnelle : fichiers → archive texte → fichiers reconstruits
Interfaces CLI et librairie Python
Respecte .gitignore et les règles d'exclusion par glob
Cons
Très faible traction (2 étoiles, 1 contributeur, 24 commits)
Modèle de permissions simplifié (utilisateur seulement, POSIX/UNIX)
Ne gère pas les fichiers non-UTF-8 ni les liens symboliques