A service can show "loaded" while producing nothing. A database can exist while being locked. A file can be present while being stale. Memory Doctor checks what matters — not just "does it exist?" but "is it actually working?"
Your AI has a WAL, LCM database, semantic search, PostgreSQL, markdown files, snapshots, flush logs, bootstrap files, and more. Any of them can fail silently. You won't know until your AI forgets something critical.
Your WAL service is registered with launchd. It shows "loaded." But it's pointing to the wrong script, producing nothing. You don't notice for days.
Your LCM database file is there. It has data. But a stale lock prevents any new writes. Summaries stop accumulating. Context compacts to nothing.
Your snapshot exists. It's the right size. But it hasn't been updated in 3 days. Your AI recovers from a 72-hour-old snapshot and misses everything since.
MEMORY.md exists and looks fine. But the "SYSTEM ARCHITECTURE" section got deleted during a bad merge. Your AI loses critical knowledge — silently.
Postgres stopped accepting connections. Your photos table, tech handoffs, dispatch jobs — all inaccessible. Your AI can't look up history.
HEARTBEAT.md got deleted. Or IDENTITY.md. Or TOOLS.md. Your AI starts up without critical configuration — and you don't know which file is missing.
Memory Doctor checks all 12 memory layers in your OpenClaw installation, color-codes the results, and tells you exactly what's broken and how to fix it.
Is session-wal.sh running? Is the WAL file fresh (last write < 10 min)? File size growing? Today's file exists? Checks the actual process AND the output — not just "loaded."
Is lcm.db accessible? Not locked? Size reasonable? Recent summaries exist? Message count growing? Can run queries successfully? Catches stale locks before they eat your context.
Is Ollama running? Is the qmd CLI available? Collection indexed? Executes a test query and verifies results. Not just "is the service up" but "can it actually find things."
Is Postgres accepting connections? Can it query your database? Do all required tables exist (photos, tech_handoffs, part_lifecycle, dispatch_jobs, agent_tasks)? Row counts for each.
File exists? Recently modified? Size > 0? Contains expected sections? Catches the "file exists but critical sections got deleted" failure mode.
Your agent's identity file. Checks existence and verifies Core Commitments and Tripwire Wisdom sections are intact. Without SOUL.md, your agent has no identity.
ops/snapshots/latest.md — exists? Fresh (< 24h)? Size reasonable? A stale snapshot means your recovery point is hours or days behind reality.
Today's memory file exists? Yesterday's? Total archive count? Daily memory files capture decisions and context changes — if they stop being created, something broke.
ops/flush-log.md — exists? Recent flush logged (< 1h)? If the flush system stops running, your AI's maximum memory loss goes from 5 minutes to infinity.
All 7 auto-loaded files present: AGENTS.md, SOUL.md, TOOLS.md, IDENTITY.md, USER.md, HEARTBEAT.md, MEMORY.md. One missing file = one blind spot.
RECOVERY.md should NOT be present (cleaned up after use). Wisdom service running? If Wisdom is down, your behavioral monitoring is blind.
ops/credentials/ directory exists with files? Without credentials, your agent can't authenticate to external services.
Run the full 12-layer check. Color-coded report with ✅/⚠️/❌ for every check. Exit code tells you the overall health. One command, instant answer.
Pass --fix and Memory Doctor attempts auto-repair: restart WAL, clear LCM locks, create missing daily memory files, clean up stale recovery files. All repairs logged.
Pass --watch for continuous monitoring. Re-checks every 5 minutes. Alerts only on state changes — when something breaks or recovers. Background it and forget.
Pass --json for machine-readable output. Integrate with dashboards, heartbeat scripts, launchd services, or your own monitoring stack.
This tool was born from experience. A WAL service showed "loaded" for a week while pointing to the wrong script. A Jarvis Doctor only checked "loaded" not "working." Memory Doctor checks what matters — actual output, actual freshness, actual health.
12 memory layers checked. Color-coded results. Auto-repair. Continuous monitoring. $95, once, forever.
Buy Now — $9514-day money-back guarantee. No questions asked.