# 2026-06-12 Daily Log (continued from 1.md)

## 04:00-06:02 UTC — WSL OpenClaw Fix Marathon

### Topology Clarification
- Larry has TWO OpenClaw instances:
  - **VPS (me/Hermes)**: `100.105.27.25`, port 18789, Telegram bot `513847…t6E8` = buckalewBot
  - **WSL local**: `100.109.176.73`, port 18789, Telegram bot `856457…O1Ew` = @Openclaw1976_bot
- Windows native OpenClaw = **disabled** (scheduled task off, no process running)
- WSL bound to loopback only (`bind: loopback`, `auth.mode=none`)
- Windows Tailscale IP `100.109.12.41` = no SSH server

### What Was Fixed on WSL
1. **Model**: `openrouter/owl-alpha` (401) → `ollama/qwen3:8b` (streaming timeout) → `ollama/llama3.2:3b` (local, fast, no rate limits)
2. **Tool profile**: `coding` → `full` (coding profile was stripping message/nodes/tts tools)
3. **active-memory plugin**: disabled (was timing out at 30s on every message, blocking the main agent)
4. **thinkingDefault**: `medium` → removed (llama3.2:3b doesn't need it)
5. **baseUrl**: `http://100.105.27.25:11434` (VPS Ollama) — still points here, but Tailscale relay issues prevent connectivity

### Current WSL Config (saved on disk)
- `agents.list[0].model = "ollama/llama3.2:3b"`
- `agents.defaults.model = "ollama/llama3.2:3b"`
- `tools.profile = "full"`
- `plugins.entries.active-memory.enabled = false`
- `models.providers.ollama.baseUrl = "http://100.105.27.25:11434"`
- `models.providers.ollama.models` includes llama3.2:3b, qwen3:8b, minimax-m3:cloud, gemma4:e2b

### Ongoing Issues
- **Tailscale Miami relay blocking TCP**: SSH and HTTP to `100.109.176.73` from VPS times out. Tailscale control-plane works (peer shows "active") but TCP CONNECT fails. `tailscale ssh` also returns `502 Bad Gateway`.
- **Windows has 29 local Ollama models** (gemma4:12b, qwen3.5, deepseek-v4-pro, etc.) — could be used from WSL via `host.docker.internal:11434` (WSL VirtualSwitch, no Tailscale needed)
- **Eden AI API key** (`eyJhbG…sY5E`): valid but **no credits** on account

### TODO for Tomorrow
1. Check if Tailscale relay is fixed (might need Windows restart)
2. If relay still broken: point WSL at Windows Ollama via `http://host.docker.internal:11434` (bypasses Tailscale entirely)
3. Restart WSL gateway: `wsl -d Ubuntu -e bash -c "systemctl --user restart openclaw-gateway"`
4. Test OpenClaw-Helper_bot on Telegram
5. If working: re-enable active-memory plugin with shorter timeout or disable recall-heavy prompt style

### Larry's Preferences Reminder
- "stop pinging me with automated reports, keep conversations between me and the agents when I initiate them"
- Wants both VPS and WSL running separately
- WSL = OpenClaw-Helper_bot (Telegram), VPS = buckalewBot (Telegram)

### Obsidian Memory Dump (00:00 UTC Jun 13)
- **Nightly dump** (`/root/scripts/nightly-obsidian-dump.sh`): ✅ Ran manually
- **Memory dump** (`/usr/local/bin/obsidian-memory-dump.sh`): ✅ Runs at 4 AM UTC, pure shell
- Dump location: `/var/www/hermes_output/obsidian/memory/`

### Shared Brain Sync (restored at 00:00 UTC)
- Tailscale direct connection restored (was relay-only for ~10h)
- `sync-shared.sh` and `sync-memories.sh` both passing
- VPS ↔ WSL bidirectional sync working

### VPS Model (CURRENT as of Jun 13)
- **Default**: `ollama/llama3.2:3b` (local, 2GB, no rate limits)
- **Avoid**: minimax-m3:cloud (weekly quota exhausted), openrouter/owl-alpha (401)
- **Gateway bind**: `loopback` (127.0.0.1:18789) — cron tools require this