What problem does it solve?
Two-layer memory architecture helps logging and recalling CEO decisions by separating raw transcripts from approved conclusions. This separation enables robust audit trails, safer decision playback, and clean recall for future sessions.
Core Features & Use Cases
- Layer 1 — Raw Transcripts: Location: memory/consultations/YYYY-MM-DD-raw.md, captures full agent contributions, critique, synthesis, and rejected arguments; not auto-loaded.
- Layer 2 — Approved Decisions: Location: memory/consultations/decisions.md; stores only CEO-approved decisions and action items; loaded automatically at the start of every advisory session and never deleted.
- Conflict Detection & Safety Guards: Checks for DO_NOT_RESURFACE violations, topic contradictions, and owner conflicts; surfaces conflicts for CEO resolution.
- Logging Workflow: From transcription to decision logging, including conflict checks and archival.
- Session Start Protocol & Proactive Triggers: On session start, loads Layer 2 decisions and flags overdue items; proactive prompts when decisions stall.
- File Structure: memory/consultations/ with decisions.md and per-session raw transcripts and archive folders.
Quick Start
Log a CEO decision after a consult and review overdue actions at the next session.