Start a conversation
The model-free dialogue layer has starter phrases, recent-turn context and selected follow-up workflows. It does not perform unrestricted language generation.
How are you?Teach a private reply
Use the Framework Teach conversation form or this chat sentence. The exact phrasing is saved with a reusable reply. Reusing it does not call a model.
When I say "daily focus", reply "What is the most useful result we can finish today?"Correct an answer
Immediately after a conversational answer, provide the preferred response. Corrections are private and versioned. They cannot replace a computed tool result.
Better reply: Let’s choose one useful result and finish it.Personalize from private memory
A memory placeholder reads only the current workspace. If the value is missing, HuLaHo asks instead of inventing it.
teach reply: {"name":"personal_greeting","examples":["greet me by name"],"reply":"Hello {memory:name}!"}Review lessons
Use replies, inspect reply, rollback reply and forget reply. Export includes active private lessons. Shared lesson changes are handled by the operator CLI.
repliesUse a teacher only when needed
The optional operator-run teacher calls an existing loopback llama.cpp chat endpoint with a supplied topic. It reads no private conversations and does not run on each chat request. Draft lessons require validation and operator review before shared activation.
Understand the boundary
Storing response patterns is not training model weights or acquiring unrestricted language understanding. Matcher tests check phrasing, not truth. Unfamiliar meanings need supported tools, sources, new lessons or clarification.
Read the methods and limitations or try the examples in the workspace.