---

Why this exists: Running Claude, Kimi, Hermes, and OpenClaw on one Mac Mini M4 with 32GB RAM is a resource management nightmare. This skill captures the hard-earned knowledge of keeping them from stepping on each other.

name: nerua1-workspace description: Complete workspace knowledge base for nerua1's multi-agent AI setup. Hardware, software stack, active issues, skills inventory, and agent responsibilities. type: reference created: 2026-04-18 authors: [nerua1, Claude Code, Kimi K2.5] version: "2.0.0"

nerua1 Workspace 🧠🦞⚡

Hardware

• Mac: Mac Mini M4, 32GB RAM, macOS 26.4
• Storage: 2TB APFS external (/Volumes/2TB_APFS/)
• Network: Standard, no VPN

Software Stack

Component	Version	Path / URL	Status
Hermes Agent	0.10.0	~/.hermes/	✅ Custom provider patched
OpenClaw	2026.4.15	/opt/homebrew/bin/openclaw	⚠️ ACP sleeps when busy
Claude Code	latest	~/.claude/	✅ Native ACP support
Kimi CLI	latest	~/.kimi/	✅ Native ACP support
LM Studio	latest	local API @ 127.0.0.1:1234/v1	⚠️ 18 models loaded, RAM tight
acpx	0.5.3	via npx acpx	✅ Hermes config override active

---

Model Aliases (Hermes)

Alias	Model	RAM Need	Use When
main	qwen3.5-27b-uncensored	~18GB	>14GB free RAM
fallback	qwen3.5-9b-uncensored	~10GB	<14GB free RAM
fast	qwen3.5-4b-uncensored	~5GB	Quick tasks
coder	qwen3-coder-30b	~22GB	Code-heavy
heavy	qwen3.5-35b	~24GB	Complex reasoning
gemma	gemma-4-31b-it	~20GB	General

---

Active ACP Sessions

Session ID	Agent	Status
aa97b585-f1e1-4e8f-85ac-ebe9720a0bea	hermes (test-session)	✅ Active
0c590567-5d56-4b88-9ebc-1cd3f94cb187	openclaw (oc-test)	⚠️ Created but not responding

---

Critical Config Files

~/.acpx/config.json                          # acpx agent overrides
~/.hermes/config.yaml                        # Hermes main config
~/.local/bin/hermes                          # RAM-aware wrapper
~/.local/bin/ram-orchestrator               # RAM monitor + ACP switcher

---

Shared Memory

• Obsidian Vault: /Volumes/2TB_APFS/openclaw-data/workspace/obsidian-memory/
• Bridge Handoff: /Volumes/2TB_APFS/openclaw-data/workspace/obsidian-memory/bridge/AGENT-HANDOFF.md
• Memory Palace: ~/.mempalace/palace (ChromaDB)

---

Active PRs / Issues

#	Repo	Title	Status
#11865	NousResearch/hermes-agent	fix(cli): named custom providers + live models	🟡 Open

---

Skills Inventory

Skill	GitHub	Published	Type
acp-bridge	nerua1/acp-bridge	✅ v1.2.0	reference
publishing-guide	nerua1/publishing-guide	✅ v1.0.0	reference
github-promotion	(this file)	📝 Draft	reference
ram-orchestrator	(local only)	📝 Draft	tool

---

Agent Specializations

Agent	Best At	Avoid
Claude Code	Architecture, planning, code review, PRs	Long context without compression
Hermes	Tool-calling, coding, terminal operations	Running when RAM < 3GB
OpenClaw	Gateway, multi-model, WhatsApp bridge	ACP commands during heavy tasks
Kimi	Fast responses, Chinese/English bilingual	Complex tool chains

---

Quick Commands

# Check RAM
vm_stat | awk '/Pages free/{f=$3}/Pages inactive/{i=$3}/Pages speculative/{s=$3}/Pages purgeable/{p=$3}END{gsub(/\./,"",f);gsub(/\./,"",i);gsub(/\./,"",s);gsub(/\./,"",p);print (f+i+s+p)*4096/1024/1024 " MB"}'

# Hermes ACP via acpx
npx acpx hermes prompt "message" --session test-session

# Switch Hermes model via ACP
npx acpx hermes prompt "/model fast" --session test-session

# RAM orchestrator (one-shot)
~/.local/bin/ram-orchestrator --threshold 4000 --model fast

# List loaded LM Studio models
curl -s http://127.0.0.1:1234/v1/models | python3 -c "import sys,json;[print(m['id']) for m in json.load(sys.stdin)['data']]"

---

PayPal Link (all repos)

If this saved you time: [☕ PayPal.me/nerudek](https://www.paypal.me/nerudek)

---

Notes

• Never run git commit/push/rebase without user confirmation
• All new public repos must include PayPal link
• Check gh repo list before publishing (anti-duplicate rule)
• Two local skills ready for publication: lm-studio-orchestrator, system-bridge

---

If this saved you time: ☕ PayPal.me/nerudek