We write autonomous AI agents, compile at the speed of light, and deploy globally in milliseconds. Yet, at the core of our OS, we still organize data using a digital metaphor from 1964: the hierarchical directory tree.
You are forced to memorize arbitrary pathways (/src/utils/auth/keys) and exact filenames just to retrieve your own code. grep and find are band-aids applied to a fundamentally broken paradigm.
You do not care where a file is. You care what it means.
That is why I built OmitFS—a production-grade, 100% local, zero-dependency semantic file system written entirely in Rust.
Search your files the way you think about them:
You type | OmitFS finds instantly |
"the production API config"
| config/prod.env
|
"meeting notes from last sprint"
| 2024-03-15_standup.docx
|
"calculus assignment"
| Math_HW_Chapter4_Final.pdf
|
No cloud. No API limits. Forever offline.
If you are building infrastructure for developers, it cannot be a fragile Python wrapper around an OpenAI API call. It needs to be fast, memory-safe, and uncompromisingly secure.
I built OmitFS completely in Rust. It compiles down to a single binary with zero background Docker containers or external databases required.
Here is the internal pipeline:
Code snippet
graph TD
A["Files dropped into raw/"] --> B{OmitFS Daemon}
B --> C["Text Extraction (PDF, DOCX, Code)"]
C --> J["Chunker (200-word windows)"]
J --> K{"SHA-256 Manifest (Skip unchanged)"}
K -->|Changed| L["EmbeddingEngine (all-MiniLM-L6-v2)"]
L --> O[("LanceDB Vector Database")]
O --> P{Two-Stage Search}
P --> Q["1. ANN Vector Search (Cosine)"]
Q --> R["2. BM25 Re-ranker (TF-IDF)"]
R --> S["Results"]
The Core Stack:
The FUSE Kernel Bridge: Using the fuser crate, OmitFS intercepts standard POSIX system calls. Virtual folders mirror standard UNIX permissions and byte-sizes, ensuring zero friction with existing CLI tools like vim, cat, and grep.
The Local Inference Engine: Powered by the Rust ML framework candle, OmitFS runs a quantized embedding model (all-MiniLM-L6-v2) locally on your CPU/GPU. It generates 768-dimensional vectors from text payloads in ~5ms.
The Vector Store: LanceDB runs natively within the Rust process. It executes sub-millisecond Approximate Nearest Neighbor (ANN) searches, followed by a BM25 keyword re-ranker for absolute precision.
The Extractor: Natively extracts text from .pdf, .docx, .xlsx, and source code. If Tesseract or Whisper are on your $PATH, it seamlessly OCRs images and transcribes audio.
Beyond Search: RAG and MCP Integration
A semantic file system is powerful, but it is also the perfect foundational memory layer for local AI workflows.
1. Ask AI (Local RAG)
By hooking OmitFS up to a local LLM runner like Ollama, you can perform Retrieval-Augmented Generation (RAG) over your entire hard drive without passing your data to the cloud.
Bash
$ omitfs ask "What integral technique did I use in problem 4?"
OmitFS retrieves the exact mathematical chunk from your PDF, feeds it to llama3 or phi4 as context, and streams the answer back to your terminal.
2. Native MCP Integration (Cursor / Claude)
OmitFS ships with native Model Context Protocol (MCP) server capabilities. You can instantly plug it into Claude Desktop or Cursor IDE. Your AI agent can autonomously query OmitFS to find the exact documentation or architectural decisions hidden across your local machine to give you better code generations.
The Benchmarks
Rust ensures OmitFS stays entirely out of your way. Because of the SHA-256 manifest, the daemon never re-indexes files that haven't changed.
Operation | CPU (i7-12th gen) | GPU (RTX 3080) |
Embed one chunk (~200 words) | ~25 ms | ~2 ms |
Index 1,000 text files | ~4 min | ~20 sec |
Search (query → results) | ~60 ms | ~10 ms |
RAG answer (Llama3 8B) | ~8 sec | ~0.8 sec |
Daemon restart (unchanged) | instant | instant |
One-Line Install
The installation is completely frictionless. A single command downloads the pre-built binary, patches your PATH, and sets up the daemon.
macOS / Linux:
Bash
curl -sSf https://raw.githubusercontent.com/Panav-Payappagoudar/OmitFS/main/install.sh | sh
Windows PowerShell:
PowerShell
irm https://raw.githubusercontent.com/Panav-Payappagoudar/OmitFS/main/install.ps1 | iex
Join the Resistance
We spend too much time managing the "containers" of our data rather than interacting with the substance itself. OmitFS strips away the UI clutter, the nested folders, and the mental overhead.
If you are a Rustacean, an AI engineer, or just someone tired of looking for lost files, I would love your feedback.
Check out the full source code on GitHub:
Panav-Payappagoudar/OmitFS on GitHub
If this project saves you time, dropping a ⭐ on the repo helps massively.