Auto Refinement
LanMate periodically reviews its own conversations, distills what worked into experience, and reuses it next time — the longer you use it, the better it fits how you work.
What Gets Refined
| Type | Content | Example |
|---|---|---|
| Prompt | Instruction patterns effective for a task class | “For weekly reports, ask about the audience first” |
| Memory | User preferences and project context | “Director Zhang’s reports use a three-part structure” |
| Skill | Methodology for repeated workflows | “Contract review: check breach clauses first” |
| Sub-agent | Dispatching experience | “Data cleaning suits the small-model tier” |
Automatic and Manual
- Auto refinement (on by default): reviews periodically during a conversation; frequency and cooldown adjustable in settings
- Manual trigger: just say “review what we learned this time” in chat
Two Layers of Experience
| Layer | Scope | Entry |
|---|---|---|
| Task-level | Current task only | Summary at the top of the task page |
| Cross-session | Shared by all tasks | Settings → Personalization → Refinement records |
Both layers can be deleted anytime; deleting an entry stops the agent from citing it, without touching conversation history.
Turning It Off
Settings → Personalization → Auto refinement; off means only manual triggers run. Turn it off when you don’t want the agent remembering anything.