| name | live-memory-stats |
| description | Show the Live Memory plugin's current status for this workspace — agent state, context-window fill, retained observations and Q&A pairs, session token cost, and the pending question queue. Use when the user asks how Live Memory is doing, how full its context is, how much it has cost, or what it has accumulated. |
Live Memory — Status
Live Memory is a persistent, LLM-backed codebase companion contributed by the
live-memory plugin. It observes the files Shofer edits and reads (and external
edits, via a filesystem watch), accumulates a per-workspace knowledge log, and
answers investigative questions through the ask_live_memory tool.
Where the status comes from
You do not need to run anything special — the current status is already surfaced
in two places:
- The system prompt. The plugin appends a
Live Memory section to the prompt
each turn (via transformSystemPrompt). It shows: whether the memory LLM is
ready (AI granted and billed-calls consent given), the active model label,
the number of retained observations and Q&A pairs, and — once the memory agent
has run at least one question — the context-window fill (current / max tokens
and a nearly-full marker).
- The chat panel (the plugin's
sidebar-panel UI). It streams the live agent
state header (Standby / Ready / Busy / Error), the context-usage bar, and the
full typed conversation (text / reasoning / tool_call parts).
What the numbers mean
| Field | Meaning |
|---|
| State | Standby (no agent yet) · Ready · Busy (a question is in flight) · Error |
| Observations | Recent file-activity markers retained (capped by maxObservations). |
| Q&A retained | Recent question/answer pairs retained (capped by maxQuestions). |
| Context | Estimated tokens in the memory agent's context window vs. its budget (maxContextTokens); flagged nearly-full past contextFillThreshold. |
| Cost | Running session estimate (USD) accumulated across ask_live_memory calls. |
| Queue | Pending questions waiting on the single-in-flight serializer. |
To read the accumulated knowledge itself, ask a question with the
ask_live_memory tool rather than re-reading files yourself.
