Owner handoff / live infrastructure reference

OpenClaw, local LLM, and SEO operations

A complete reference for the Windows Ollama fleet, private VPS API bridge, OpenClaw gateway, SEO skill activation, monitoring, recovery paths, and remaining work.

Verified 2026-08-06 04:19 UTC (2026-08-06 00:19 Toronto). Credentials and private keys are deliberately omitted. Download the Markdown handoff.

15Models ready

128OpenClaw skills ready

23 / 27MCP healthy

376MCP tools exposed

85%NS3 root disk used

Architecture

The model API stays private. Windows hosts inference and telemetry; the VPS loopback bridge is the only route consumed by OpenClaw and the production SEO service.

01 / InferenceWindows Ollama

i7-2600, 16 GB RAM, GTX 745 4 GB. Models live in E:\Ollama\models.

02 / MeterTelemetry proxy

OpenAI-compatible API on Windows 127.0.0.1:11438, with token accounting.

03 / TunnelReverse SSH

VPS 11436 maps to the model API; VPS 11437 maps to status.

04 / AgentsOpenClaw + SEO

Gateway and FairChoice SEO consume http://127.0.0.1:11436/v1.

Bridge verified. OpenClaw, model API, tunnel, token meter, and Ollama all returned healthy status. The private /v1/models route returned HTTP 200.

Model fleet

Fifteen models are tracked. All 15 are installed, API-live, and registered in OpenClaw. Keep only one model loaded at a time during inference on this hardware.

State	Model	Size	Role
Ready	`phi4-mini-openclaw:latest`	2.5 GB	Primary fast agent and SEO model
Ready	`phi4-mini:latest`	2.5 GB	Fast local fallback
Ready	`llama3.2:3b`	2.0 GB	Lightweight fallback
Ready	`qwen2.5-coder:7b`	4.7 GB	Coding fallback
Ready	`ministral3:8b-instruct-2512-q4_k_m`	5.2 GB	Tools and general agents
Ready	`Hermes-3-Llama-3.1-8B Q4_K_M`	4.9 GB	Function calling and JSON
Ready	`Granite-3.3-8B-Instruct Q4_K_M`	4.6 GB	Structured and RAG work
Ready	`DeepSeek-R1-Distill-Llama-8B Q4_K_M`	4.6 GB	Reasoning specialist
Ready	`gemma4-coder-fable5:12b-q4_k_m`	6.87 GB	Larger coding specialist
Ready	`Ornith-1.0-9B Q4_K_M`	5.6 GB	Agentic coding
Ready	`LFM2.5-8B-A1B Q4_K_M`	5.16 GB	Fast tool calling
Ready	`phi4-mini-reasoning:latest`	3.2 GB	Fast local reasoning
Ready	`phi4-reasoning:14b-q4_K_M`	11.0 GB	Deeper reasoning and code
Ready	`qwen3:14b`	9.3 GB	General reasoning and coding
Ready	`gemma3:12b`	8.1 GB	Image and text understanding

Parallel completion finished. The six unfinished pulls completed with independent retries, all 15 models are available through the OpenAI-compatible API, and the live fleet page is authoritative.

SEO activation

The free local LLM now backs the production FairChoice SEO dashboard. Collision-prone skills were namespaced, and the previously broken drift monitor was made self-contained.

Completed

Production SEO runtime corrected to use phi4-mini-openclaw:latest with phi4-mini:latest fallback.
agentic-seo activated under a unique name.
Thirteen ultimate-seo-* skills activated.
seo-drift-monitor repaired and verified against FairChoice.
All 15 additions allowlisted for seo-orchestrator.

Verified state

128 ready skills for the SEO orchestrator.
98 repositories tracked; 68 installed or active.
22 of 22 monitored services up on the live SEO status feed.
23 of 27 MCP servers healthy, exposing 376 tools.
Drift check: 0 critical, 0 warning, 0 info.

Service ownership note. Production is the active gunicorn service on 127.0.0.1:18880. Port 5000 is also alive from an unmanaged root-session Python process while its systemd unit is inactive. Keep 18880 as production and choose one owner for the legacy UI later.

2026-08-06 repair update. The old :8613/seo-repos/ URL is restored directly on NS3 through nginx. The moved media-mdla container is running on 162.35.188.122, and NS3 now writes live status.json locally first, then mirrors status.json and mcp-health.json into 162's media-mdla copy. The dashboard's MCP filename matcher was corrected so local files such as mdla_hands_server.py show as configured through their OpenClaw server IDs.

Restored SEO reference	Status	Notes
Main SEO inventory	200	Live services feed: 22/22 up.
API Keys reference	200	Copied back from 162 media-mdla.
Fable tools	200	Copied back from 162 media-mdla.
AI Research tools	200	Copied back from 162 media-mdla.
Install report	200	Copied back from 162 media-mdla.
Audit report	200	Copied back from 162 media-mdla.
Work report	200	Copied back from 162 media-mdla.

Pending MCP	Boundary
`agentql-mcp`	AgentQL API key
`brightdata-mcp`	BrightData API token
`gsc-mcp`	Google service-account credentials
`serpapi-mcp`	SerpApi key and stdio-to-HTTP correction

Public references

Current public entry points. The VPN response is protected by design.

OpenClawHTTP 200 / gateway UI LLM fleetHTTP 200 / live status API SEO inventoryHTTP 200 / reports and health Provider dashboardHTTP 200 / token and provider state LLM chatHTTP 200 / model chat VPN dashboardHTTP 401 / expected authentication

Recovery runbook

Use the narrowest command that restores the failed layer. Avoid restarting OpenClaw for static report edits.

Windows: start the full fleet

cd C:\Worker\openclaw-phi-local
.\start-llm-fleet-services.ps1

Restore only the tunnel

cd C:\Worker\openclaw-phi-local
.\start-openclaw-phi-tunnel.ps1

VPS: verify core services

systemctl status openclaw-gateway --no-pager
curl -sS http://127.0.0.1:11436/v1/models
systemctl status fairchoice-seo-dashboard --no-pager
        openclaw skills list --agent seo-orchestrator
systemctl start mcp-health.service

Completion verified. The Windows updater is restart-safe but idle: all tracked models are installed, the private API bridge is healthy, and OpenClaw registration is complete.

Risks and priorities

The system is operating, but the disk ceiling and legacy process ownership should be handled before expanding the stack.

Watch: NS3 root disk is 85% full. The emergency 91% condition has improved, but keep reclaiming space before adding another Docker stack or large crawl artifact to NS3.

Next actions

Audit disk usage and reclaim safe space without deleting live images.
Choose one owner for the legacy SEO UI on port 5000.
Test one OpenClaw SEO prompt and confirm token telemetry increments.
Keep the LLM watchdog and reverse tunnel supervised; no downloader resume is currently needed.
Add external SEO credentials only when the accounts exist.

Do not do

Do not expose the private model endpoint publicly without authentication.
Do not place credentials in this report or dashboard source.
Do not re-enable duplicate services on occupied ports.
Do not install another heavy browser stack while NS3 root disk is still above 80%.
Do not assume “installed” means “registered in OpenClaw.”

Files and backups

Restore individual files only. Do not roll back unrelated changes made by another operator.

Windows project
C:\Worker\openclaw-phi-local

SEO deployment source
C:\Worker\seo-repos-activation

OpenClaw config
/root/.openclaw/openclaw.json

SEO production source
/srv/openclaw/fairchoice/seo-agent

SEO skills
/root/.openclaw/workspace/skills

SEO site on NS3
/var/www/seo-repos served at http://192.227.221.215:8613/seo-repos/

Migrated media-mdla copy on 162
162.35.188.122 media-mdla:/var/www/seo-repos, public through http://162.35.188.122:8613/seo-repos/

Activation backup
/root/_repair_backups_seo_activation_20260802T232901Z

OpenClaw config backup
/root/.openclaw/openclaw.json.bak-seo-activation-20260802T233313Z

Dashboard backup
/root/seo-dashboard-backup-20260802T234950Z in media-mdla

Runtime-env backup
/root/_repair_backups_seo_handoff_20260803T005958Z