How Your Knowledge Fits Together

Suppose your project will launch on the web first because a small team cannot maintain mobile apps yet. That choice and its reason make one useful memory.

A memory can also be a fact, preference, lesson, or warning. Keep enough context that someone reading it next month can understand when it applies.

During a session, you might discuss launch options, reject a few ideas, and choose an approach. The whole conversation is not equally useful later.

Saving the final decision gives your next conversation something specific to retrieve. Do not assume that every message in every connected tool has been saved.

At the end of that session, a handoff can record the decision, the work completed, and the remaining task—such as drafting the launch checklist.

The Space’s Brief gives the next session a starting point. It is different from search: read the Brief to get oriented, then recall a particular detail when you need it.

Later, you might save the launch date, supported browsers, and release checklist. A Wiki page can bring those findings together into a launch plan with citations.

Open a citation to inspect the material behind a claim. If a tracked source changes in Wenlan, dependent pages can be refreshed. Whether the new text is applied directly or waits for approval depends on the page’s ownership and review lock.

The desktop app lets you browse and edit the library. Connected tools such as Claude Code and Codex can search it and use what they find in a conversation.

Underneath, a local service—the daemon—stores the data and handles retrieval. MCP is the connection that lets an AI tool talk to that service. You do not need to understand the server architecture to save your first memory.

Wiki pages and session notes have readable Markdown files. A local database also keeps memories, links, indexes, and update state.

For a full backup or migration, preserve the documented data directory rather than copying only the Markdown pages.

Saving and searching do not require you to configure a separate language model. Your connected AI can help write memories and organize pages during a conversation.

Automatic background organization and page synthesis need their assigned models. You can choose local or supported remote providers; the provider choice affects where the material is processed.