Summary
The answer you half-remember is in a conversation from three weeks ago.
It is built around not flooding the context: search returns snippets, conversations are fetched in parts, and scope is limited before results come back. The command search is the sharpest of them - the invocation you used once and never wrote down, found by what it did rather than by its name.
What it is
A search server over two local stores: Memex's conversation history and the project directories in your workspace. It returns targeted snippets and scoped results rather than whole files.
What you get
- Conversation search by text, metadata and filters, with relevance scoring
- A specific part of a conversation, retrieved without pulling the whole thing into context
- Project search by content, file type and name
- A project summary with the technologies it detects
- Commands, CLI usage and code snippets found in past conversations, filtered by the kind of command you mean
- Faceted filtering on dates, projects and file types
Requirements
A local Memex install - the server reads the history directory under Application Support and the projects in your workspace folder, both at fixed paths. Node, and a build from the repository.