Shareable configuration for OpenClaw — memory system, skills, and agent instructions.
A battle-tested configuration for running OpenClaw as a personal AI assistant:
- Three-tier memory architecture — Daily logs, curated long-term memory, and searchable deep knowledge
- Semantic memory search — Vector embeddings via LM Studio (local, free) or OpenAI for finding relevant context
- Decision-making frameworks — Bezos's one-way/two-way doors, certainty thresholds, priority filters
- Group chat behavior — When to speak, when to stay silent, how to react like a human
- Ready-to-use skills — Limitless Pendant, Fireflies.ai, and Quo phone integrations
Set up openclaw-config from https://github.com/TechNickAI/openclaw-config
Update my openclaw config
| File | Purpose |
|---|---|
AGENTS.md |
Complete operating instructions (~400 lines) |
SOUL.md |
AI personality template |
USER.md |
Human profile template |
TOOLS.md |
Local environment notes |
HEARTBEAT.md |
Periodic check config |
IDENTITY.md |
Quick reference card |
Tier 1: Always-Loaded
MEMORY.md— Curated essentials (~100 lines max)
Tier 2: Daily Context
memory/YYYY-MM-DD.md— Today + yesterday's logs
Tier 3: Deep Knowledge (Semantic Search)
memory/people/,memory/projects/,memory/topics/,memory/decisions/- Indexed with vector embeddings for semantic similarity search
- Use
memory_search("query")to find relevant context - Supports LM Studio (local, recommended) or OpenAI for embeddings
Skills are standalone UV scripts — Python with inline dependencies, no project setup required.
| Skill | What it does |
|---|---|
| parallel | Web search & content extraction via Parallel.ai (Python, tested) |
| limitless | Query Limitless Pendant lifelogs |
| fireflies | Search Fireflies.ai meeting transcripts |
| quo | Access Quo business phone calls & texts |
| openclaw | Manage openclaw-config installation & updates |
Workflows are autonomous agents that run on a schedule. Unlike skills (single-purpose tools), workflows have state, learn over time, and maintain their own rules and logs.
| Workflow | What it does |
|---|---|
| email-steward | Manages your inbox — archives, deletes, alerts on urgent |
Each workflow has:
AGENT.md— The algorithm (updates with openclaw-config)rules.md— Your preferences (you customize, never overwritten)agent_notes.md— Learned patterns (grows over time)logs/— Execution history
uv run --with pytest pytest tests/ -vIntegration tests auto-skip if API keys aren't set.
Your AI uses these to decide what to remember long-term:
- Durability — Will this matter in 30+ days?
- Uniqueness — Is this new or already captured?
- Retrievability — Will I want to recall this later?
- Authority — Is this reliable?
- File-based memory — Text files beat databases for AI context
- Markdown over JSON — Language models work better with prose
- Two-way doors — Act freely on reversible decisions; ask first on irreversible ones
- Quality over quantity — In groups, participate, don't dominate
PRs welcome! Keep templates generic (no personal content).
MIT
Author: TechNickAI