\r \r

Hermes Memory Sync\r

\r Automated daily memory extraction from Hermes Agent session logs. Reads session JSON/JSONL files, groups conversations by day, extracts key topics/decisions/tools used, and writes structured summaries to workspace/memory/YYYY-MM-DD.md.\r \r

When to Use\r

\r

• Setting up memory persistence for Hermes Agent (beyond built-in memory/session_search)\r
• Recovering or backfilling memory from earlier sessions\r
• Creating a human-readable daily log of what was discussed\r
• Auditing session history by topic or tool usage\r \r

Installation\r

\r

1. Place the script\r

\r Copy hermes-memory-sync.py to your workspace:\r \r

# Windows\r
copy hermes-memory-sync.py C:\path	o\workspace\\r
\r
# Linux/macOS\r
cp hermes-memory-sync.py ~/workspace/\r
```\r
\r
### 2. Verify\r
\r
```bash\r
cd /path/to/workspace\r
python hermes-memory-sync.py stats\r
```\r
\r
Expected output: shows active days, total messages, sessions, and existing memory files.\r
\r
## Usage\r
\r
```bash\r
# Show coverage gaps between sessions and existing memory files\r
python hermes-memory-sync.py compare\r
\r
# Backfill today's memory\r
python hermes-memory-sync.py backfill today\r
\r
# Backfill a specific date\r
python hermes-memory-sync.py backfill 2026-05-03\r
\r
# Backfill all missing dates (first-time setup)\r
python hermes-memory-sync.py backfill all\r
\r
# Show stats\r
python hermes-memory-sync.py stats\r
```\r
\r
## Cron Setup (Hermes Native)\r
\r
Use Hermes' built-in cron to run daily at 3 AM:\r
\r
```\r
cronjob action=create name="memory-sync-daily"\r
  prompt="Run Hermes memory sync: cd /path/to/workspace && python hermes-memory-sync.py backfill today"\r
  schedule="0 3 * * *"\r
```\r
\r
The cron job creates files at `workspace/memory/YYYY-MM-DD.md` each morning.\r
\r
## Output Format\r
\r
Each memory file contains:\r
\r
```\r
# 📅 YYYY-MM-DD\r
\r
**会话数:** N | **消息总数:** N\r
**用户提问:** N | **助手回复:** N | **工具调用:** N\r
**使用的模型:** model1, model2\r
\r
## 🎯 讨论主题\r
- Topic 1\r
- Topic 2\r
\r
## 💬 关键对话\r
**Q:** User question...\r
> **A:** Assistant response...\r
\r
## ⚡ 决策/方案\r
- Decision item...\r
\r
## 🛠️ 工具使用\r
- Tool call summary...\r
\r
---\r
*自动生成于 YYYY-MM-DD HH:MM，来自 N 个会话*\r
```\r
\r
## Supported Session Files\r
\r
The script reads two types of Hermes session data:\r
\r
### 1. `session_*.json` (full session records)\r
- Located at `%LOCALAPPDATA%/hermes/sessions/` (Windows)\r
- Complete conversation history with `messages` array\r
- Includes `session_id`, `model`, `platform`, `session_start`\r
\r
### 2. `YYYYMMDD_HHMMSS_*.jsonl` (per-message logs)\r
- Same sessions directory\r
- One JSON object per line with `{role, content, timestamp}`\r
- Uses Hermes format (NOT OpenClaw format)\r
\r
## Pitfalls\r
\r
- **Memory usage** — periodically review and consolidate memory entries. Over-full memory causes truncation and lost context.\r
- **JSONL vs JSON confusion** — The sessions directory contains both `.jsonl` and `.json` files. The script handles both, but `request_dump_*.json` files (individual request/response dumps) are intentionally skipped to avoid duplication.\r
- **Don't confuse with ClawHub `memory-sync`** — That skill is designed for OpenClaw's JSONL format (`{type, message}` schema). This Hermes-native implementation directly parses Hermes format (`{role, content}` schema). See `clawhub-skills-install` skill for format differences.\r