diff --git a/README.md b/README.md
index 7b79be5..999168e 100644
--- a/README.md
+++ b/README.md
@@ -1,15 +1,103 @@
 # PsiRepacker
-Repacking NT hash files from Troy Hunt to binary format for Get-Badpasswords solution
 
-# Usage
-Simply run the executable with two parameters:
+Repacking NT hash files from Troy Hunt's [Pwned Passwords](https://haveibeenpwned.com/Passwords) to compact binary format for the [Get-Badpasswords](https://github.com/yourlink) solution.
+
+## Versions
+
+### PyPsiRepacker (Python — recommended)
+
+A streaming Python implementation that converts NTLM hash text files to binary with **near-zero memory usage**. Unlike the C++ version, it does not load all entries into RAM — it processes line-by-line, relying on the input already being sorted (as produced by [PwnedPasswordsDownloader](https://github.com/HaveIBeenPwned/PwnedPasswordsDownloader)).
+
+**Requirements:** Python 3.6+ (no external dependencies)
+
+**RAM usage:** ~0 (streams line by line)
+
+#### Prerequisites (Windows)
+
+1. **Install Python 3.6 or later** — download from [python.org](https://www.python.org/downloads/)
+   - During installation, check **"Add Python to PATH"**
+2. **Verify** by opening Command Prompt or PowerShell:
+   ```
+   python --version
+   ```
+3. **Clone or download** this repository:
+   ```
+   git clone https://github.com/improsec/PsiRepacker.git
+   cd PsiRepacker
+   ```
+
+#### Usage
+
+Run from the repository root:
 
 ```
-PsiRepacker <input filepath> <output filepath>
+python -m pypsirepacker <input filepath> <output filepath>
+```
+
+**Windows example (Command Prompt or PowerShell):**
+
+```
+python -m pypsirepacker "D:\output\hashes\pwnedpasswords_ntlm.txt" "D:\output\bin\pwnedpasswords_ntlm.bin"
 ```
 
-For example:
+**Linux / macOS example:**
 
 ```
-PsiRepacker "C:\pwned-passwords.txt" "C:\pwned-passwords.bin"
+python3 -m pypsirepacker /data/pwnedpasswords_ntlm.txt /data/pwnedpasswords_ntlm.bin
 ```
+
+#### Options
+
+| Flag | Description |
+|------|-------------|
+| `--no-verify` | Skip sort-order verification (not recommended) |
+
+Sort-order verification is enabled by default. If a hash is found out of order, conversion aborts immediately — this means the input was not pre-sorted and a streaming approach cannot produce a valid binary file.
+
+#### Example run (Pwned Passwords v10, ~2 billion NTLM hashes)
+
+| | File | Size |
+|---|---|---|
+| **Input** | `pwnedpasswords_ntlm.txt` | ~70 GB |
+| **Output** | `pwnedpasswords_ntlm.bin` | ~30.6 GB |
+
+```
+Counting entries in pwnedpasswords_ntlm.txt...
+Found 2,052,742,897 entries.
+Converting to binary: pwnedpasswords_ntlm.bin
+   1,800,000,000 / 2,052,742,897 (87.7%)
+   1,850,000,000 / 2,052,742,897 (90.1%)
+   1,900,000,000 / 2,052,742,897 (92.6%)
+   1,950,000,000 / 2,052,742,897 (95.0%)
+   2,000,000,000 / 2,052,742,897 (97.4%)
+   2,050,000,000 / 2,052,742,897 (99.9%)
+Done. Wrote 2,052,742,897 entries (32,843,886,352 bytes + 8 byte header).
+Completed in 870.2 seconds.
+```
+
+### PsiRepacker (C++ — original)
+
+The original C++ implementation loads all hashes into memory, sorts them, and writes the binary output. This requires **~50 GB RAM** for a full Pwned Passwords NTLM file (~1.7 billion entries) and is Windows-only.
+
+```
+PsiRepacker <input filepath> <output filepath>
+```
+
+Use this version if your input file is **not pre-sorted** and you have sufficient RAM.
+
+## Binary output format
+
+The output `.bin` file has the following structure:
+
+| Offset | Size | Description |
+|--------|------|-------------|
+| 0 | 8 bytes | Entry count (uint64, little-endian) |
+| 8 | 16 bytes each | NTLM hashes packed as raw binary, sorted |
+
+Each 32-character hex NTLM hash (e.g. `A4F49C406510BDCAB6824EE7C30FD852`) is stored as 16 raw bytes — halving the size of the text representation and enabling fast binary search.
+
+## Workflow
+
+1. Download NTLM hashes using [PwnedPasswordsDownloader](https://github.com/HaveIBeenPwned/PwnedPasswordsDownloader) with the `-n` (NTLM) flag
+2. Run PsiRepacker (Python or C++) to convert to binary
+3. Use the `.bin` file with Get-Badpasswords to audit Active Directory passwords
diff --git a/pypsirepacker/__init__.py b/pypsirepacker/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/pypsirepacker/__main__.py b/pypsirepacker/__main__.py
new file mode 100644
index 0000000..3508538
--- /dev/null
+++ b/pypsirepacker/__main__.py
@@ -0,0 +1,46 @@
+"""
+Command-line entry point for PyPsiRepacker.
+
+Usage:
+    python -m pypsirepacker <input_path> <output_path> [--no-verify]
+"""
+
+import argparse
+import sys
+import time
+
+from .repacker import repack
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        prog="pypsirepacker",
+        description=(
+            "Convert Troy Hunt Pwned Passwords NTLM hash files from text to "
+            "compact binary format for Get-Badpasswords."
+        ),
+    )
+    parser.add_argument("input", help="Path to NTLM hash text file (HASH:count per line)")
+    parser.add_argument("output", help="Path for output binary file")
+    parser.add_argument(
+        "--no-verify",
+        action="store_true",
+        help="Skip sort-order verification (not recommended)",
+    )
+
+    args = parser.parse_args()
+
+    start = time.perf_counter()
+
+    try:
+        count = repack(args.input, args.output, verify_sort=not args.no_verify)
+    except (FileNotFoundError, ValueError) as e:
+        print(f"Error: {e}", file=sys.stderr)
+        sys.exit(1)
+
+    elapsed = time.perf_counter() - start
+    print(f"Completed in {elapsed:.1f} seconds.")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/pypsirepacker/repacker.py b/pypsirepacker/repacker.py
new file mode 100644
index 0000000..9f8e8bd
--- /dev/null
+++ b/pypsirepacker/repacker.py
@@ -0,0 +1,114 @@
+"""
+Streaming NTLM hash repacker for Troy Hunt's Pwned Passwords lists.
+
+Converts text-based NTLM hash files (HASH:count format) to compact binary
+format for use with Get-Badpasswords. Processes line-by-line with near-zero
+memory usage, relying on the input being pre-sorted (as produced by
+PwnedPasswordsDownloader).
+"""
+
+import struct
+import sys
+import os
+
+
+HASH_HEX_LEN = 32
+HASH_BIN_LEN = 16
+HEADER_SIZE = 8  # uint64 little-endian
+
+
+def count_lines(filepath: str) -> int:
+    """Count lines in a text file efficiently using a raw byte buffer."""
+    count = 0
+    buf_size = 1024 * 1024  # 1 MB chunks
+    with open(filepath, "rb") as f:
+        while True:
+            buf = f.read(buf_size)
+            if not buf:
+                break
+            count += buf.count(b"\n")
+    return count
+
+
+def repack(input_path: str, output_path: str, *, verify_sort: bool = True) -> int:
+    """
+    Convert a Troy Hunt NTLM hash text file to sorted binary format.
+
+    The input file must have lines in the format: <32-char NTLM hash>:<count>
+    The output is a binary file: 8-byte entry count (uint64 LE) followed by
+    packed 16-byte hash entries.
+
+    Args:
+        input_path:   Path to the NTLM text file (e.g. pwned-passwords-ntlm.txt).
+        output_path:  Path for the output binary file.
+        verify_sort:  If True, verify that input hashes are in sorted order and
+                      abort if not. Default True.
+
+    Returns:
+        Number of entries written.
+
+    Raises:
+        FileNotFoundError: If input file does not exist.
+        ValueError:        If the file format is invalid or sort order is broken.
+    """
+    if not os.path.isfile(input_path):
+        raise FileNotFoundError(f"Input file not found: {input_path}")
+
+    # Validate format by peeking at first line
+    with open(input_path, "r") as f:
+        first_line = f.readline().strip()
+
+    if len(first_line) < HASH_HEX_LEN or first_line[HASH_HEX_LEN] != ":":
+        raise ValueError(
+            f"Not a valid Troy Hunt NTLM hash file. "
+            f"Expected format: [32-char NTLM hash]:[count]. "
+            f"Got: {first_line[:50]}"
+        )
+
+    # Count entries for the binary header
+    print(f"Counting entries in {input_path}...")
+    total = count_lines(input_path)
+    print(f"Found {total:,} entries.")
+
+    # Stream convert
+    print(f"Converting to binary: {output_path}")
+    written = 0
+    prev_hash = None
+
+    with open(input_path, "r") as fin, open(output_path, "wb") as fout:
+        fout.write(struct.pack("<Q", total))
+
+        for line_num, line in enumerate(fin, 1):
+            line = line.strip()
+            if not line:
+                continue
+
+            hash_hex = line[:HASH_HEX_LEN].upper()
+
+            if verify_sort and prev_hash is not None and hash_hex < prev_hash:
+                # Clean up partial output
+                fout.close()
+                os.remove(output_path)
+                raise ValueError(
+                    f"Sort order violation at line {line_num:,}: "
+                    f"{hash_hex} < {prev_hash}. "
+                    f"Input file is not sorted. Cannot proceed with streaming conversion."
+                )
+
+            prev_hash = hash_hex
+
+            try:
+                fout.write(bytes.fromhex(hash_hex))
+            except ValueError:
+                raise ValueError(
+                    f"Invalid hex at line {line_num:,}: {hash_hex}"
+                )
+
+            written += 1
+
+            if written % 50_000_000 == 0:
+                pct = (written / total) * 100 if total else 0
+                print(f"  {written:>14,} / {total:,} ({pct:.1f}%)")
+
+    print(f"Done. Wrote {written:,} entries ({written * HASH_BIN_LEN:,} bytes + {HEADER_SIZE} byte header).")
+    return written