GreenBubbles

GreenBubbles connects an existing coding agent to your own WeChat history on a Mac. It reads the database files already maintained by WeChat, so you do not need to export every conversation before searching it. The project provides a command-line interface and an agent skill for turning selected messages into notes with references to the original conversation.

Key features

• Browse conversations, search message text, and retrieve messages with available attachment paths.
• Use an existing agent to organize selected conversations into Markdown notes, with source messages attached to each fact.
• Update notes incrementally rather than rewriting the entire archive each time.
• Create encrypted backups protected by a recovery phrase when you explicitly request a backup.

The repository describes the project as a research alpha. It requires Apple silicon and macOS 14 or later. Public builds read history and cannot send messages. The project uses the MIT license and is independent of Tencent.

A practical workflow

Start by reading the official setup and usage guide. Review the database-key setup before installing: it needs administrator access and re-signs your local WeChat copy. Once configured, run greenbubbles source status to check access, then greenbubbles chats to identify the intended conversation. Search a narrow topic with greenbubbles messages search --conversation "Alice" --query "project". Replace the sample name and keyword with your own selection.

Give your agent the supplied personal-memory skill and ask it to summarize only that conversation. Review the message citations, remove unnecessary personal details, and save the result in your existing note workflow. Use a small sample before requesting a month of history.

Pricing and limitations

The code is open source; this listing does not verify a paid product plan. An existing coding-agent subscription or API can carry its own charges. The optional built-in summarizer requires a Gemini API key and is billed separately by Google.

Local storage does not mean every AI workflow stays local. A cloud coding agent sends the messages it reads to its provider. The skill is an instruction file, not an access-control boundary. WeChat's private database format can change, and compatibility needs to be checked after updates. Protect notes, database keys, backups, and recovery phrases as personal data.

FAQ and alternatives

Does it send WeChat messages? Public builds cannot send messages.

Is it a hosted chatbot? It is a local data bridge that works with an agent you already use.

Where should the notes go? Obsidian is a related option for managing Markdown notes. Claude Code and Codex CLI are agent choices, not replacements for the WeChat reader.

Explore the productivity category for adjacent tools. The next step is to review the official key-setup documentation and test one selected conversation before expanding access.