Infrastructure Migration Summary - 2026-01-23

Purpose: Complete guide for infrastructure migration to NAS - includes Docker services, git repositories, worktrees, and wikis

Status: ✅ MIGRATION COMPLETE - 100% NAS-Based Infrastructure

---

WHAT GOT MOVED

1. Git Infrastructure Moved to NAS (2026-01-23)

All Repositories (67 GitLab Projects)

• Status: ✅ COMPLETE - All repos are bare repos on NAS
• Location: /Volumes/AgentPlatform/repos/bare/blueflyio/
• Size: 9.4GB total
• Organization:
  • Top-level repos: gitlab_components.git, platform-agents.git, security-policies.git
  • Agent-platform group: /Volumes/AgentPlatform/repos/bare/blueflyio/agent-platform/[project].git
  • OSSA group: /Volumes/AgentPlatform/repos/bare/blueflyio/ossa/
• Local Action: NO local clones exist or needed - all work via NAS worktrees

All Worktrees (NAS-Centralized Strategy)

• Status: ✅ COMPLETE - 100% NAS-based worktree workflow
• Location: /Volumes/AgentPlatform/worktrees/[DEVICE]/[DATE]/[PROJECT]/[BRANCH]/
• Device Namespaces:
  • shared/ - Multi-device work (Mac M3, M4, code-server, phone, iPad)
  • m3/ - Mac M3 performance-specific work
  • m4/ - Mac M4 performance-specific work
• Old Local Worktrees: ✅ REMOVED
  • Cleaned up: ~/Sites/blueflyio/.worktrees/ (all old local worktrees removed)
  • Freed: ~500MB
  • All bare repos pruned to remove stale registrations
• Benefits: Work from ANY device - same files, same state, always in sync

All Wikis (Documentation)

• Status: ✅ COMPLETE - All wikis on NAS only
• Location: /Volumes/AgentPlatform/wikis/blueflyio/
• Wikis Migrated:
  • technical-docs.wiki/ - Platform documentation (already had GitLab origin)
  • api_normalization.wiki/ - API Normalization project docs
• Old Local Wikis: ✅ REMOVED
  • Cleaned up: ~/Sites/blueflyio/_WIKI/ (entire directory removed)
  • Freed: 2.7GB (including old .zip backups)
  • No symlinks or local copies remain
• Access: Direct git operations in NAS wiki directories (no worktrees needed for wikis)

Infrastructure Alignment Complete

• AGENTS.md: ✅ Updated to NAS-centralized workflow (v1.1.0)
• CLAUDE.md: ✅ Updated with NAS worktree paths and device namespaces
• Wiki Docs: ✅ Updated separation-of-duties.md, system-overview.md, platform-overview.md
• Architecture Decisions: ✅ worktree-strategy-nas-centralized.md marked APPROVED
• Local Setup Required: MINIMAL
  • ~/Sites/blueflyio/CLAUDE.md (instructions)
  • ~/Sites/blueflyio/.claude/ (config)
  • Mount NAS: /Volumes/AgentPlatform/
  • NO repos, NO worktrees, NO wikis locally

---

2. Services Moved to NAS (blueflynas.tailcf98b3.ts.net)

Agent-Tracer (Knowledge Graph API)

• Status: Deployed on NAS K3s cluster
• Location: development namespace, port 3007
• Pod: agent-tracer-67df67fc46-q9kzh
• Local Docker: Can be removed (postgres-tracer, redis-tracer, agent-tracer containers)

KAGENT Platform (23 Agents)

• Status: Running in K3s kagent namespace
• Agents: Assessment, Onboarding, Renewal, ROI Calculator, K8s agents, Helm agents, etc.
• API/UI: KAGENT API, UI, and controller all on NAS
• Local Docker: Can be cleaned up (old KAGENT images)

Phoenix Observability

• Status: Running in K3s monitoring namespace
• Local Docker: Old Phoenix images can be removed

3. Services Moved to Vast.ai

LLMLingua Compression Service

• Status: Deployed on Vast.ai RTX 4090
• Instance ID: 30386215
• SSH: ssh -p 26214 root@ssh9.vast.ai
• Cost: $0.29/hr ($207/month)
• Purpose: Token compression (79x reduction for large contexts)
• Local Docker: Never existed locally (deployed directly to Vast.ai)

---

WHAT TO DO ON YOUR OTHER MAC

Phase 0: Git Infrastructure Setup (REQUIRED FIRST)

NAS-centralized workflow means NOTHING local except instructions:

1. Mount NAS (Finder: ⌘K):

   # Mount via NFS
   nfs://192.168.68.54/volume1/AgentPlatform
   # Or via Tailscale:
   nfs://blueflynas.tailcf98b3.ts.net/volume1/AgentPlatform
   
   # Verify mount
   ls /Volumes/AgentPlatform/
   # Should show: repos/, worktrees/, wikis/

2. Create minimal local directory:

   mkdir -p ~/Sites/blueflyio/.claude
   cd ~/Sites/blueflyio/

3. Copy CLAUDE.md from NAS:

   # Option A: From wiki
   cp /Volumes/AgentPlatform/wikis/blueflyio/technical-docs.wiki/getting-started/CLAUDE.md ~/Sites/blueflyio/
   
   # Option B: From this Mac (via Tailscale)
   scp bluefly.tailcf98b3.ts.net:~/Sites/blueflyio/CLAUDE.md ~/Sites/blueflyio/

4. That's it! Everything else (repos, worktrees, wikis) is on NAS.

Complete setup guide: /Volumes/AgentPlatform/wikis/blueflyio/technical-docs.wiki/getting-started/nas-setup-guide.md

Also see: /tmp/other-computer-setup.md (quick reference created for Mac M3)

---

Phase 1: Docker Cleanup (Safe - Do This Second)

Run this single command to remove all unused Docker resources:

docker system prune -af --volumes

This removes:

• Dangling images (untagged, unused)
• Stopped containers
• Build cache
• Unused volumes
• Unused networks

Expected recovery: 20-30 GB (varies by Mac)

What it does NOT remove:

• Running containers
• Images used by running containers
• Volumes attached to running containers

---

Phase 2: Service-Specific Cleanup (After Verification)

A. Agent-Tracer Local Services

Verify it's on NAS first:

# Confirm it's running on NAS (ask primary Mac user to verify)
kubectl get pods -n development | grep agent-tracer

If confirmed on NAS, remove local:

# Stop and remove local agent-tracer stack
docker ps | grep tracer  # Should show nothing after NAS deployment

# If any tracer containers exist locally:
docker stop $(docker ps -q --filter "name=tracer")
docker rm $(docker ps -aq --filter "name=tracer")

# Remove tracer images
docker images | grep tracer | awk '{print $3}' | xargs docker rmi -f

B. Old Phoenix Images

Keep only latest, remove old versions:

# List Phoenix images
docker images arizephoenix/phoenix

# Remove all but the latest (they're running on NAS now)
docker images arizephoenix/phoenix --format "{{.ID}} {{.CreatedAt}}" | tail -n +2 | awk '{print $1}' | xargs docker rmi -f

Expected recovery: ~5 GB

C. Old LibreChat Images

# Remove old versions
docker images ghcr.io/danny-avila/librechat | grep '<none>' | awk '{print $3}' | xargs docker rmi -f

Expected recovery: ~4-5 GB

D. Old ClickHouse Images

docker images clickhouse/clickhouse-server | grep '<none>' | awk '{print $3}' | xargs docker rmi -f

Expected recovery: ~1-2 GB

E. Old Prometheus Images

docker images prom/prometheus | grep '<none>' | awk '{print $3}' | xargs docker rmi -f

Expected recovery: ~1 GB

---

Phase 3: GitLab Runner Cache Cleanup (If Applicable)

Check if GitLab runners are active:

docker ps | grep gitlab-runner

If NO runners active, clean up cache volumes:

docker volume ls | grep runner- | awk '{print $2}' | xargs docker volume rm

Expected recovery: 1-2 GB

---

Phase 4: Old Compose Stack Volumes (Selective)

Only remove if you've confirmed these services are on NAS/K3s:

CSMA Platform Volumes (if on K3s)

docker volume rm csma-platform_grafana-data \
  csma-platform_langflow-data \
  csma-platform_loki-data \
  csma-platform_neo4j-data \
  csma-platform_neo4j-logs \
  csma-platform_phoenix-data \
  csma-platform_postgres-data \
  csma-platform_prometheus-data \
  csma-platform_redis-data

Old Infrastructure Volumes

docker volume rm infrastructure_librechat_logs \
  infrastructure_librechat_meilisearch_data \
  infrastructure_librechat_mongo_data \
  infrastructure_librechat_storage

Expected recovery: 5-10 GB

---

WHAT TO KEEP LOCALLY

Active Docker Compose Stacks (DO NOT REMOVE)

1. agent-router Stack

• Status: Keep for local development/monitoring
• Services: Grafana, Prometheus, Postgres, Redis, Qdrant, Mongo (10 services)
• Purpose: LLM Platform infrastructure

2. agent-mesh Stack

• Status: Keep for local KAGENT development
• Services: postgres, redis, qdrant (3 services)
• Purpose: KAGENT local services

3. DDEV Stacks (If Active)

• ddev-drupal-demo - Keep if actively used for demos
• ddev-ssh-agent - Keep (DDEV infrastructure)

---

SERVICES STATUS REFERENCE

Running on NAS (K3s Cluster)

Running on Vast.ai

Deprecated Locally (Safe to Remove)

• Agent-tracer local containers
• Old Phoenix images
• Old LibreChat images
• Old ClickHouse images
• Old Prometheus images
• GitLab runner caches (if no active runners)
• CSMA platform volumes (if on K3s)

---

COMPLETE CLEANUP COMMANDS

⚠️ FILE POLICY NOTE: The commands below were previously packaged as a .sh script, but .sh scripts are restricted per project FILE POLICY. Run these commands manually or via npm scripts instead.

Run these commands manually after verification:

# Complete Docker cleanup after migration to NAS and Vast.ai

set -e

echo "=== Phase 1: Remove Unused Resources ==="
docker system prune -af --volumes
echo "✅ Phase 1 Complete"

echo ""
echo "=== Phase 2: Remove Old Service Images ==="

# Phoenix
docker images arizephoenix/phoenix --format "{{.ID}}" | tail -n +2 | xargs -r docker rmi -f

# LibreChat
docker images ghcr.io/danny-avila/librechat | grep '<none>' | awk '{print $3}' | xargs -r docker rmi -f

# ClickHouse
docker images clickhouse/clickhouse-server | grep '<none>' | awk '{print $3}' | xargs -r docker rmi -f

# Prometheus
docker images prom/prometheus | grep '<none>' | awk '{print $3}' | xargs -r docker rmi -f

echo "✅ Phase 2 Complete"

echo ""
echo "=== Phase 3: Remove GitLab Runner Caches ==="
docker volume ls --format "{{.Name}}" | grep runner- | xargs -r docker volume rm
echo "✅ Phase 3 Complete"

echo ""
echo "=== Phase 4: Remove Old Stack Volumes ==="

# CSMA Platform (now on K3s)
docker volume rm -f csma-platform_grafana-data \
  csma-platform_langflow-data \
  csma-platform_loki-data \
  csma-platform_neo4j-data \
  csma-platform_neo4j-logs \
  csma-platform_phoenix-data \
  csma-platform_postgres-data \
  csma-platform_prometheus-data \
  csma-platform_redis-data 2>/dev/null || true

# Old infrastructure
docker volume rm -f infrastructure_librechat_logs \
  infrastructure_librechat_meilisearch_data \
  infrastructure_librechat_mongo_data \
  infrastructure_librechat_storage 2>/dev/null || true

echo "✅ Phase 4 Complete"

echo ""
echo "=== Cleanup Summary ==="
docker system df
echo ""
echo "✅ All cleanup complete!"

Usage: Copy and paste the commands above into your terminal, or package them in an npm script in package.json (preferred over .sh files).

---

VERIFICATION STEPS

1. Check What's Running Locally

docker ps

2. Check Disk Usage Before/After

docker system df

3. Verify Services on NAS (From Primary Mac)

# Check K3s pods
kubectl get pods -n development
kubectl get pods -n kagent
kubectl get pods -n monitoring

# Check NAS services
curl http://blueflynas.tailcf98b3.ts.net:3007/health

4. Verify LLMLingua on Vast.ai

ssh -p 26214 root@ssh9.vast.ai 'curl -s http://localhost:8000/health'

---

EXPECTED DISK RECOVERY

---

TIMELINE

1. Immediate (Safe): Run Phase 1 cleanup → ~20-30 GB recovery (5 minutes)
2. After NAS Verification: Run Phase 2-4 cleanup → ~15-25 GB recovery (10 minutes)
3. Total Active Work: 15-20 minutes
4. Total Recovery: 35-55 GB

---

SUPPORT

Questions? Ask the primary Mac user (thomas.scola) to verify:

• Services running on NAS K3s cluster
• Services running on Vast.ai
• What's safe to remove locally

NAS Access:

• Host: blueflynas.tailcf98b3.ts.net
• User: bluefly
• Via Tailscale VPN

Vast.ai Access:

• Instance: 30386215
• SSH: ssh -p 26214 root@ssh9.vast.ai

---

SUMMARY

What Changed:

Git Infrastructure (2026-01-23):

• ✅ All 67 repositories → NAS bare repos (/Volumes/AgentPlatform/repos/bare/blueflyio/)
• ✅ All worktrees → NAS device namespaces (/Volumes/AgentPlatform/worktrees/[DEVICE]/)
• ✅ All wikis → NAS only (/Volumes/AgentPlatform/wikis/blueflyio/)
• ✅ Local worktrees removed (~/Sites/blueflyio/.worktrees/ - freed 500MB)
• ✅ Local wikis removed (~/Sites/blueflyio/_WIKI/ - freed 2.7GB)
• ✅ NAS-centralized worktree strategy APPROVED and DEPLOYED

Docker Services (2026-01-22):

• ✅ Agent-tracer → NAS K3s
• ✅ KAGENT Platform → NAS K3s
• ✅ Phoenix → NAS K3s
• ✅ LLMLingua → Vast.ai RTX 4090

What to Do on Other Mac:

1. Phase 0: Mount NAS, create minimal local setup (REQUIRED FIRST)
2. Phase 1: Run Docker cleanup immediately (safe, 20-30GB)
3. Phase 2-4: Run Docker service cleanup after NAS verification (15-25GB)
4. Total Recovery: 35-55 GB of disk space

What to Keep Locally:

• ~/Sites/blueflyio/CLAUDE.md (instructions only)
• ~/Sites/blueflyio/.claude/ (config)
• Docker: agent-router stack, agent-mesh stack, active DDEV stacks
• NO repos, NO worktrees, NO wikis - all on NAS

Benefits:

• Work from ANY device (Mac M3, M4, code-server, phone, iPad)
• Same files, same state, always in sync
• Infrastructure independence achieved

---

Generated: 2026-01-23 Status: ✅ MIGRATION COMPLETE - 100% NAS-Based Infrastructure For: Complete infrastructure migration guide (git + Docker) Reference: See /Volumes/AgentPlatform/wikis/blueflyio/technical-docs.wiki/getting-started/nas-setup-guide.md