Apex stack build — live progress

Auto-updates as criteria land. Hard guardrails at the bottom — what I will NOT do without Mark's go.
CRITERION 1
Off-machine TP3 backup (Postgres + MinIO)
Postgres dump ✓ — 552 MB, sha256 09bc3fad08dc67ff2061c41ec0725c6678eec5d55d8eb93ee04370b5f382b395 MinIO tar-pipe (4 non-audio buckets): tp3-biometric-raw-parquet 597 MB ✓ tp3-legacy-archive 690 KB ✓ tp3-omi-raw-events 276 MB (~27% of 1 GB) tp3-omi-dead-letter pending Audio archive (7 GB) deferred — regeneratable from OMI source devices.
RUNNING
CRITERION 2
Local models pulled into tp3_ollama
qwen3:30b-a3b ✓ 18 GB mistral-small3.2:24b ✓ 15 GB qwen3:14b ✓ 9.3 GB GPU verified: qwen3:14b 49 tok/s; qwen3:30b-a3b 35 tok/s (MoE)
✓ DONE
CRITERION 3
LiteLLM up, 5 models routed
Docker pull stayed blocked even after full Desktop restart (credential helper session bound to user logon, can't refresh without Mark logout/login). PIVOTED to pip install on Apex host Python 3.11. - TP3 LiteLLM Proxy scheduled task running, port 4000 localhost - /v1/models HTTP 200, returns: claude-sonnet, claude-opus, local-big (qwen3:30b-a3b), local-fallback (mistral 24b), local-floor (qwen3:14b) - LITELLM_MASTER_KEY generated + in .env - Auto-restart on failure (RestartCount=3)
✓ DONE
CRITERION 4
Open WebUI reachable via Tailscale
Installed via pip (same Docker pivot). TP3 Open WebUI scheduled task registered + running. First-launch downloading 30 HF embedding-model files on init (slow, no HF_TOKEN). Once downloads finish, port 3000 will serve. http://100.88.195.118:3000 from your phone via Tailscale.
RUNNING
CRITERION 5
"Local First" + "Claude" presets working
awaits criteria 3+4
PENDING
CRITERION 6
Zero TP3 regression check
tp3_memories_local: 118,993 rows (was 118,519 baseline) — fresh ingest continuing. Last 1h: omi_webhook 355, omi 3, phone_call 2. All 17 containers up + healthy pre-restart. Re-checks after Docker Desktop restart.
✓ DONE
CRITERION 7
Live dashboard at reports.thebarnetts.info
this page
✓ DONE
CRITERION 8
Agent-coordination protocol pushed
memory file: feedback_agent_coordination_protocol_2026-06-04.md twin-memory memory_id: 230 git: 129c0d7
✓ DONE
CRITERION 9
DC trip questions out of active eval
golden_questions.json total=18, active=13, trip_archive=5 run_eval.py: default excludes *_archive categories
✓ DONE
CRITERION 10
Phase 0 maintenance scripts staged (NOT RUN)
C:\Users\Breezy\apex_stack_build\maintenance\ contents: README.md (DO NOT RUN until Mark home) litellm_config.yaml litellm_up.ps1 openwebui_up.ps1 Order documented: 0a WSL config / 0b DB cred rotation / 0c CF+Make+NVIDIA keys → then LiteLLM up → then Open WebUI.
✓ DONE

Hard guardrails — never crossed

Last update: (this page bakes the timestamp at render — refresh to see latest).
Goal target: Friday June 6 by 6 PM ET when Mark lands.
Build runbook: git show origin/claude/apex-agent-stack-3ztfV:project_apex_agent_stack_runbook_2026-06-01.md on the memory repo.