convrtr
Start converting

2 September 2026

Unpacking Ren'Py RPA Archives in the Browser: Byte Structure & Extraction

Ren'Py is the dominant open-source visual novel game engine, powering thousands of games on Steam, Itch.io, and mobile. If you inspect the game/ folder of almost any Ren'Py title, you will find files named archive.rpa, scripts.rpa, or audio.rpa.

These are RPA (Ren'Py Archive) files. While they serve a similar purpose to ZIP archives, double-clicking them in your operating system will fail. They are deliberately designed with an obfuscated index to prevent accidental spoilers and raw script browsing.

This guide explains the exact byte structure of RPA-2.0 and RPA-3.0 archives, how the index table is obfuscated with XOR keys, and how you can unpack them directly into a standard ZIP file in your browser using the RPA to ZIP tool.

The RPA Header Signatures

An RPA archive starts with a human-readable ASCII header line ending in a newline (0x0A). Ren'Py has evolved across several archive revisions:

  1. RPA-3.0: Starts with RPA-3.0 followed by an 8 or 16-hexadecimal character offset, a space, and an 8 or 16-hexadecimal character key: 
    RPA-3.0 012a4f00 42c89f10
  2. RPA-2.0: Starts with RPA-2.0 followed by an 8-character hex index offset, followed by the index block: 
    RPA-2.0 00fa3210
  3. RPA-1.0: A legacy format where the index sits directly at the start of the archive without obfuscation.

The key values in the header define two critical pieces of information:

  • Index Offset: The exact byte position in the file where the archive's file directory table begins.
  • XOR Key: A 32-bit or 64-bit mask used to de-obfuscate the index block.

Reading the Obfuscated Directory Index

Unlike ZIP files which store an uncompressed Central Directory Record at the end of the file, RPA archives place a zlib-compressed or raw byte stream containing a serialized Python dictionary at the index offset.

In RPA-3.0:

  1. Seek to the offset decoded from the ASCII header.
  2. Read the remaining byte stream.
  3. Apply the key across the index payload using bitwise XOR (byte ^ key[i % key_len]).
  4. Decompress the resulting bytes using zlib inflate.
  5. Parse the dictionary containing filenames, byte offsets, lengths, and optional prefixes.

Each record inside the decompressed index contains:

  • filename: The relative virtual path of the asset (e.g. images/character_happy.png or audio/bgm01.ogg).
  • offset: The absolute byte position where the raw file data starts within the .rpa file.
  • length: The exact size of the payload in bytes.
  • prefix: An optional byte sequence prepended to file contents in memory.

Browser-Based In-Memory Extraction

Historically, extracting RPA files required installing Python and running external command-line scripts like rpatool or unrpa.

Because Ren'Py archives simply concatenate assets at continuous byte offsets, our parser reads the file into a streaming buffer:

  1. Validates the RPA-3.0 or RPA-2.0 header.
  2. Unmasks the dictionary index in memory.
  3. Iterates over each file entry and slices the exact byte ranges from the underlying Uint8Array.
  4. Repackages all slices into a standard RFC 1951 Deflate container.
  5. Generates an uncompressed or compressed .zip archive ready for instant download.

Because everything executes in a dedicated Web Worker on your device, gigabyte-sized archives unpack in seconds without uploading a single byte to an external server. You can try it now with convrtr's RPA to ZIP extractor.