10 The teach Workflow

The teach subcommand of the CLI takes a plain-text file — autobiography draft, journal, notes, anything narrative — and uses an LLM to extract durable third-person facts, then submits them as a batch of memories. It’s the fastest way to seed your store with the content of a document you’ve already written.

This is the only subcommand that calls a generative LLM directly. The server is not involved in the extraction; the CLI talks to the chat-completion endpoint on its own. The extracted facts go into the server via the same remember_batch path any other client uses.