OpenClaw Mega Cheatsheet 2026 — Every CLI Command, Config Option & Workspace File

April 21, 2026 · ClawReady Team

MoltFounders published an OpenClaw Mega Cheatsheet today — a dense, structured reference covering the full CLI surface, channel setup, workspace files, memory system, model auth, sessions, and slash commands in one place. If you've ever wished the docs were scannable rather than narrative, this is the closest thing to it.

Here's a distillation of what's in it and why each section matters.

Quick start (in case you've forgotten the right flags)

# Install
npm install -g openclaw@latest

# Onboard (guided, installs daemon)
openclaw onboard --install-daemon

# Or minimal
openclaw onboard --flow quickstart --skip-channels

# Start gateway
openclaw gateway --port 18789

The --mode local vs --mode remote flag distinction is worth knowing: local mode runs the gateway on the same machine, remote mode connects to an already-running gateway on another device or server.

Channel setup — the commands that actually work

Channel	Method	Command
WhatsApp	QR Scan	`openclaw channels login`
Telegram	Bot Token	`openclaw channels add --channel telegram --token $TOKEN`
Discord	Bot Token	`openclaw channels add --channel discord --token $TOKEN`
Signal	Linked Device	`openclaw channels add --channel signal`
Slack	Bot Token	`openclaw channels add --channel slack`
Google Chat	Service Account	`openclaw channels add --channel googlechat`
MS Teams	Bot Registration	`openclaw channels add --channel msteams`
iMessage	macOS Native	macOS bridge (no CLI flag)

Diagnostics: openclaw channels status --probe to test live connectivity. openclaw channels logs --channel [id] for per-channel log tailing.

Workspace files — what each one does

These live at ~/.openclaw/workspace/ (or your configured workspace root):

File	Purpose
`AGENTS.md`	Operating instructions — what the agent does, how it delegates
`SOUL.md`	Persona, tone, hard boundaries
`USER.md`	User info, preferences, timezone, how to address them
`IDENTITY.md`	Agent name, emoji, theme
`MEMORY.md`	Curated long-term facts (DM sessions only)
`memory/YYYY-MM-DD.md`	Daily append-only log — auto-read at session start
`TOOLS.md`	Notes about local tools, environment, available CLIs
`HEARTBEAT.md`	Idle-cycle checklist — what the agent does between conversations
`BOOT.md`	Startup checklist

Root path: ~/.openclaw/workspace — everything relative to this unless overridden in openclaw.json.

Memory system — the full picture

OpenClaw has layered memory, and most users only use one layer:

Daily logs (memory/YYYY-MM-DD.md) — append-only, today + yesterday loaded at session start
Long-term memory (MEMORY.md) — curated facts, loaded only in main DM session
Vector search (memory_search tool) — semantic search across memory chunks (~400 tokens per chunk)
QMD backend (memory.backend = "qmd") — BM25 + vector + reranking; experimental but powerful
Hybrid search — default 0.7/0.3 weights (vector similarity + BM25 keyword relevance)

Provider auto-selection order: local GGUF → OpenAI → Gemini → Voyage. Set explicitly in memorySearch.provider to avoid surprises.

openclaw memory index --all   # Index entire workspace
openclaw memory search "X"    # Semantic search

Models — the commands you actually need

openclaw models list --all                        # All available models
openclaw models set anthropic/claude-opus-4-7     # Set primary
openclaw models fallbacks add openrouter/x/y      # Add fallback
openclaw models auth setup-token                  # Anthropic auth (recommended)
openclaw models auth add --provider openai        # Add API key
openclaw models status --probe                    # Live auth check
openclaw models aliases add opus anthropic/claude-opus-4-7  # Create alias

Failover cooldown tiers: 1 min → 5 min → 1 hour. If a provider fails, OpenClaw backs off exponentially before retrying.

Session config — the part most people miss

Session scope (session.dmScope) controls how much context different users share:

main — single shared session (default, fine for personal use)
per-peer — separate session per contact
per-channel-peer — separate session per channel + contact pair
per-account-channel-peer — most isolated

⚠️ The cheatsheet calls this out explicitly: use per-channel-peer for multi-user inboxes to prevent context leakage between users. If your OpenClaw handles multiple people (team use, customer service), this is not optional.

Session reset options: daily (resets at 4am local, default) or idle (sliding idle window). You can override per session type or per channel.

The honest gap this cheatsheet exposes

A reference this comprehensive is useful — but it also shows how much configuration surface OpenClaw has. Install → onboard → channel → workspace files → memory backend → model auth → session scope → fallbacks → slash commands. Each step has subtleties, and getting any one wrong produces silent failures rather than obvious errors.

That's not a criticism of OpenClaw. It's a feature: this level of configurability is why OpenClaw can serve a solo developer, a small business, and a multi-agent org on the same codebase. But it means setup matters more than most tools.

Full cheatsheet

The complete MoltFounders Mega Cheatsheet is at moltfounders.com/openclaw-mega-cheatsheet. Worth bookmarking — it covers slash commands, exec safety, sub-agent patterns, and plugin config that we didn't excerpt here.

Want your OpenClaw configured correctly from the start? ClawReady handles every layer — channel, memory, model auth, and session scope.