live-memory-config

name	live-memory-config
description	View or change the Live Memory plugin's configuration for this workspace — the provider profile the memory LLM uses, retention caps, the external-edit watch glob, the compaction interval, and the context-window budget/threshold. Use when the user asks to point Live Memory at a different model/provider, tune how much it remembers, or change its context budget.

Live Memory — Configuration

The live-memory plugin reads its settings from the standard per-plugin config surface (ctx.config, default-merged from the manifest config schema). Values are stored under pluginConfigs["live-memory"] and edited in Settings → Plugins → Live Memory, the same place the plugin is enabled and granted its billed-AI consent.

Keys

Key	Default	What it controls
profileRef	"" (host default)	The provider-profile name/id the memory LLM uses (ctx.ai.buildHandler). This is the plugin analogue of the built-in apiConfigId.
maxObservations	400	How many recent file-activity observations to retain per workspace.
maxQuestions	50	How many recent question/answer pairs to retain.
watchGlob	**/*	Glob (under the granted filesystem roots) watched for external edits.
compactIntervalMs	300000	Background compaction interval (ms); 0 disables it.
maxContextTokens	128000	The memory agent's context-window budget (tokens).
contextFillThreshold	0.8	Fraction of the budget past which the window is flagged "nearly full".

Notes

• Enabling the plugin is the on/off switch — there is no separate start/stop command. Disabling it tears down its watchers and background service.
• Billed AI requires a second, explicit "uses AI (billed)" consent beyond the plugin's permissions.ai grant; without it the memory LLM stays a denying stub and questions return a clear not-consented error rather than silently billing.
• Changing profileRef takes effect on the next ask_live_memory question (the memory agent builds its handler lazily). maxContextTokens / contextFillThreshold take effect when the agent is next (re)created.