openclaw-nextagent
# openclaw-nextagent
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
openclaw-nextagent
Custom OpenClaw setup — persistent memory with pgvector, 17 custom skills, Discord-native agent, Claude Sonnet 4.6.
A complete, production-running customization of OpenClaw deployed on a Linux laptop as a full AI assistant + content platform.
What This Is
Official OpenClaw gives you a powerful AI assistant framework. This repo shows how we took it further:
| Official OpenClaw | This Setup (NexAgent 凌骁) |
|---|---|
| Session-only memory (SQLite) | Persistent pgvector memory + structured facts |
| Generic assistant | Named persona (凌骁) with SOUL.md + IDENTITY.md |
| ClawHub skill ecosystem | 17 custom vertical skills |
| Multi-channel support | Discord-native (DM + guild moderation) |
| npm/Docker deployment | Native systemd, zero Docker |
| openclaw.json config | Multi-layer: AGENTS.md + SOUL.md + IDENTITY.md |
Architecture
Discord (DM + Guild)
│
OpenClaw Agent (凌骁 / Líng Xiāo)
│
├── AGENTS.md ← skill routing + exec policy + memory rules
├── SOUL.md ← persona baseline
├── IDENTITY.md ← name, timezone, language
│
├── Memory Service (FastAPI :18800)
│ ├── PostgreSQL 17 + pgvector ← semantic vector search
│ ├── Structured facts store ← domain/key/value
│ └── Redis ← hot-knowledge cache
│
├── Business Backend (Flask :4001)
│ ├── Blog system (NexAgent AI Blog)
│ ├── Course platform (OpenClaw Club)
│ └── Discord moderation log
│
└── Frontend (Next.js :3000, nginx, Cloudflare Tunnel)
└── nextagent.ca
Repository Structure
openclaw-nextagent/
├── README.md
├── docs/
│ ├── migration-guide.md ← Full migration: official → this setup
│ ├── database-setup.md ← SQLite → PostgreSQL migration
│ └── skills-architecture.md ← How to design your own skill layers
├── workspace/
│ ├── AGENTS.md ← Template (redacted private sections)
│ ├── SOUL.md ← Persona template
│ └── IDENTITY.md ← Identity template
├── skills/
│ ├── memory-system/ ← pgvector memory skill
│ ├── self-improvement/ ← learning + hot-knowledge cache
│ ├── blog-manager/ ← blog CRUD skill (generic)
│ ├── blog-fetcher/ ← daily news → Claude writes → publish
│ ├── reminder/ ← reminders + iCal subscriptions
│ ├── task-queue/ ← async Claude Code tasks
│ ├── google-workspace/ ← Gmail/Calendar/Drive
│ ├── github-contributor/ ← Git operations
│ ├── tavily-search/ ← real-time web search
│ ├── web-search/ ← DuckDuckGo search
│ ├── cloudflare-deploy/ ← Cloudflare Tunnel deployment
│ ├── fullstack-dev/ ← end-to-end project workflow
│ └── games/ ← mini-games
├── sql/
│ ├── memory_schema.sql ← Complete memory DB schema
│ └── blog_schema.sql ← Blog posts table with pgvector
├── systemd/
│ ├── oc-api.service ← Flask backend service
│ └── openclaw-memory.service ← Memory service
└── scripts/
├── blog_fetcher/
│ ├── fetcher.py ← Multi-source news scraper
│ ├── auto_publisher.py ← Claude writes + publishes articles
│ └── openclaw_blog_writer.py ← Daily OpenClaw update analysis
└── daily_memory_audit.sh ← Nightly memory maintenance
Quick Start
1. Install OpenClaw
npm install -g openclaw
openclaw onboard # choose Anthropic, enter API key
2. Set up PostgreSQL + pgvector
sudo apt install postgresql-17 postgresql-17-pgvector
sudo -u postgres psql -c "CREATE USER db_admin WITH PASSWORD 'yourpass';"
sudo -u postgres psql -c "CREATE DATABASE my_agent_db OWNER db_admin;"
psql -h 127.0.0.1 -U db_admin -d my_agent_db -c "CREATE EXTENSION vector;"
psql -h 127.0.0.1 -U db_admin -d my_agent_db -f sql/memory_schema.sql
3. Deploy Memory Service
git clone https://github.com/YaoS-Code/agentic-memory
cd agentic-memory/memory-service
python3 -m venv venv
./venv/bin/pip install -r requirements.txt
# Configure DB connection in config.py or .env
./venv/bin/uvicorn main:app --host 127.0.0.1 --port 18800
4. Install skills
cp -r skills/ ~/.openclaw/workspace/skills/
cp workspace/AGENTS.md ~/.openclaw/workspace/AGENTS.md
cp workspace/SOUL.md ~/.openclaw/workspace/SOUL.md
5. Configure Discord
openclaw configure --section channels
# Select Discord, enter Bot Token
6. Set systemd services (optional, replaces Docker)
sudo cp systemd/*.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable oc-api openclaw-memory
sudo systemctl start oc-api openclaw-memory
Key Design Decisions
Why PostgreSQL instead of SQLite?
- Vector search: pgvector HNSW enables semantic memory search across all past conversations
- Persistence: memories survive restarts, accumulate over time with configurable decay
- Scale: concurrent connections, JSONB, GIN indexes for fast full-text search
Why Discord-only?
We run a Discord community (OpenClaw Club). Discord gives us guild management, role-based access, slash commands, and button approvals — more than enough for a team setup.
Why systemd instead of Docker?
Running on a thin laptop (7.4GB RAM). Native systemd saves ~2GB RAM vs Docker overhead, starts faster, and has simpler dependency ordering.
Why AGENTS.md instead of plugins.json?
AGENTS.md is a markdown file the LLM reads directly. It serves as routing table + permission policy + memory rules + self-learning rules in one human-readable place. No separate config format to learn.
Version Compatibility
| This setup version | OpenClaw | Claude Model |
|---|---|---|
| Current | v2026.4.1 | claude-sonnet-4-6 |
| Previous | v2026.3.28 | claude-sonnet-4-5 |
See docs/migration-guide.md for full version changelog and impact analysis. Also see the NexAgent Blog for illustrated migration write-ups.
Related
- agentic-memory — The memory service used by this setup
- OpenClaw — The underlying agent framework
- NexAgent — Where this all runs in production
License
MIT — use freely, customize to your needs.