Turn WeChat Chats into Searchable Local Archives with WeChat AI Memory

Project decisions, requirement changes, and critical commitments often live inside WeChat private messages and group chats. But when you need to feed that context to an LLM, a few verbal summaries aren’t enough – the model guesses from incomplete information. WeChat AI Memory solves this by reading Windows WeChat 4.x data locally, filtering messages by conversation, date, and keywords, and exporting to PDF, Markdown, and structured JSON – all without any data leaving your machine.

Installation and First Launch

Download WeChatAIMemory-Portable.exe from the GitHub Release page. Double‑click to run – no installation or extraction required. The portable version extracts runtime libraries to Windows temp directory on first start, so waiting 10–20 seconds is normal. If you choose the ZIP package, extract the entire archive and keep the _internal folder alongside the EXE – copying only the executable won’t work.

The program is not code‑signed yet, so Windows SmartScreen may show “unknown publisher”. Frida, used to read the currently logged‑in WeChat process, might trigger alerts from some security software. As long as you download from the official repository and verify the source, you can safely add an exclusion.

Connecting to WeChat Data

Open WeChat AI Memory – it automatically detects the account directory of Windows WeChat 4.x. Click “Connect WeChat”, then WeChat restarts. Within 5 minutes, scan the QR code or confirm login from your phone. After connection, pick a contact or group chat, select a date range via the calendar, and optionally filter by keywords.

To retrieve original images, click “Read Images”. If the program can’t fetch image configs (especially on older WeChat versions), open one chat image in WeChat first, then retry the read operation.

Important caveat: WeChat updates may change internal data formats. The current version is verified on Windows WeChat 4.x, but “the app launches” doesn’t guarantee “all WeChat versions work”. If you run into compatibility issues, include your Windows version, WeChat version, and a sanitised error log when filing an issue.

Data Stays Local – No Uploads

The core principle: nothing leaves your computer. Chat content, images, database keys, and exported files stay local. No telemetry, no remote API calls, no message sending, recall, or modification. Decryption works on a temporary database copy – the key is never written to disk, and the original WeChat database files are never altered.

Temporary databases and recovered images live in the system temp folder and are deleted when the data source closes. Speech transcription uses a local faster‑whisper model; the model and transcription cache are stored under %LOCALAPPDATA%\WeChatAIMemory. First‑time use requires downloading the ~500 MB small model – after that, it works offline.

Supported Message Types and Output Formats

Capability Implementation
Text messages Directly read and displayed
Images V2/WXGF original recovery, EXIF orientation correction, thumbnails and 300 DPI high‑resolution separate pages
Files Copied to the _assets folder alongside the export
Voice messages SILK audio read, transcribed locally with faster‑whisper, text cached
System messages Fully covered

Three export formats: PDF for archiving and sharing, Markdown for easy editing and reading, and structured JSON for programmatic processing. When generating outputs, images and file attachments are copied to a _assets directory (named after the PDF), with relative paths used in the Markdown and JSON.

Works Without WeChat Too

If you don’t have WeChat data handy or just want to test the export pipeline, select “JSON File” as the data source in the GUI. The repository includes examples/demo_chat.json with sample data; field definitions are detailed in the JSON format documentation.

The command‑line interface is flexible for scripting:

wechat-context-exporter `
  --source examples/demo_chat.json `
  --conversation eco-project `
  --start 2026-08-21 `
  --end 2026-08-22 `
  --output outputs/demo-context.pdf `
  --markdown outputs/demo-context.md `
  --json outputs/demo-context.json

The WeChat data adapter is isolated from the rendering and export modules. Even if WeChat’s format changes, the JSON data source and output pipeline remain independent – a pragmatic architectural choice.

Common Pitfalls and Solutions

SmartScreen and antivirus blocks – The app isn’t signed, and Frida’s behaviour may be flagged. Add a folder exclusion or temporarily disable real‑time protection. Always download from the official Release page, never from third‑party re‑uploads.

WeChat update breaks reading – WeChat 4.x internal formats change. If you just updated WeChat, you might need to wait for a new version of the tool. As a temporary workaround, roll back WeChat or submit a compatibility report with details.

Speech model download is slow – The first download of the ~500 MB small model may be slow on some networks. You can manually download the model and place it in %LOCALAPPDATA%\WeChatAIMemory.

Images not showing – If image configs aren’t auto‑detected (older WeChat), open one image in WeChat chat first, then click “Read Images”.

Run from Source

For developers or debugging, Python 3.11+ is required:

git clone https://github.com/LOGO127/wechat-ai-memory.git
cd wechat-ai-memory
py -3 -m venv .venv-gui
.\.venv-gui\Scripts\python.exe -m pip install -e ".[gui,dev]"
.\.venv-gui\Scripts\python.exe app.py

Or double‑click start.cmd. Full testing, packaging, and directory details are in the development documentation.

Quick Action Checklist

  1. Download WeChatAIMemory-Portable.exe from the Release page.
  2. Log in to Windows WeChat 4.x.
  3. Double‑click the executable, wait for the first‑time extraction.
  4. Click “Connect WeChat” – WeChat restarts, then scan to log in.
  5. Pick a contact or group → select date range → enter keywords (optional).
  6. Click “Read Images” if you need original pictures.
  7. Choose a save location and click “Generate Memory Archive”.
  8. Get PDF, Markdown, and JSON exports.

FAQ

Q: Can I use this as a full backup tool for WeChat chats?
It exports filtered messages by conversation, date, and keywords into PDF/Markdown/JSON. It is not an official full‑backup tool and won’t export all raw data formats.

Q: Does speech transcription require internet?
Only for the first model download (~500 MB for small). After that, it works completely offline. No voice or chat content is uploaded.

Q: Can I feed the exported JSON directly to an LLM?
Yes. The JSON contains structured message fields suitable for programmatic processing. You can read it with code and use it as context for RAG or agent workflows.

Q: What if WeChat updates and the tool stops working?
WeChat 4.x version upgrades may change internal formats. File an issue with your Windows version, WeChat version, and a sanitised error log – the developer will adapt.

Q: Does the program modify my WeChat data?
No. It only reads a temporary copy of the database. The original database is untouched, and the encryption key is never saved.

Q: What’s the difference between the portable EXE and ZIP package?
The EXE runs directly (but extracts files on first launch). The ZIP needs full extraction and starts faster, which helps with antivirus checks.

Q: Is code signing required for normal use?
Not functionally, but SmartScreen will show “unknown publisher”. Download only from the official Release and verify the source.

Q: How do I report a bug?
Use the GitHub Issues page. Do not upload real chat databases, keys, contact names, or unredacted screenshots. Provide sanitised logs and version details.