Otto — OpenClaw AI Agent Configuration

Core Architecture

Running on OpenClaw — open-source agent system with support for multiple models, channels, and tools.

Runtime:

Host: Mac mini M4, 16GB RAM
OS: macOS 26.3.1 (arm64)
Node: v25.6.0
OpenClaw: v2026.4.11

🧠 Models & Reasoning

Primary Model: `ollama/glm-5.1:cloud`

Why: Fast, responsive, good for daily tasks
Context window: 131k tokens
Cost: Free (local/Cloudflare)

Fallback Models

1. ollama/kimi-k2.5:cloud — strong reasoning, good backup

2. ollama/qwen3-coder-next:cloud — good for coding tasks

3. ollama/qwen2.5-coder:7b — local fallback (7B params)

Alternative Session: `anthropic/claude-sonnet-4-6`

Name: "Strategický" (for complex work)

Use case: More strategic/complex work, better reasoning

Context: 200k tokens

Specialized Models

gemma4:31b-cloud — image-aware tasks
nemotron-3-nano:30b-cloud — reasoning mode enabled
nemotron-3-super:cloud — reasoning mode enabled

🎤 Text-to-Speech (TTS)

Skill: sag (ElevenLabs with macOS say UX)

Provider: ElevenLabs

Use case: Voice storytelling, audiobook-style delivery

Features: mac-style say UX, multi-voice support

🖼️ Image Capabilities

Image Analysis

Model: qwen3.5:cloud (primary)
Capabilities: text + image inputs

Image Generation

Primary: google/gemini-3-pro-image-preview
Provider: Google Gemini

🎬 Video Generation

Provider: Configured (qwen/wan2.6-t2v or similar)

Capabilities: text-to-video, reference image/video, audio background

Duration: customizable (provider-dependent)

🌐 Channels & Connectivity

iMessage (Apple)

Setting	Value
Enabled	✅ YES
DM Policy	allowlist only
Group Policy	allowlist only

Setting	Value
Enabled	✅ YES
Self-chat	enabled
Debounce	disabled
Media limit	50 MB

🔌 Gateway & Service

Gateway

Setting	Value
Local port	18789
Mode	local (loopback only)
Auth	token-based
Tailscale	OFF (local only)

Services

Gateway: Running (LaunchAgent)
Node: Not installed (not needed)

⚡ Tools & Capabilities

Category	Setting	Value
Web Tools	Search	DuckDuckGo (enabled)
	Web Fetch	enabled
Agent-to-Agent	Enabled	YES
	Allow	`sessions_spawn`
Exec	Security	full (host-level)
	Ask mode	on-miss
Memory	Backend	qmd

🔑 External APIs

Service	Purpose	Status
Ollama	Local LLM hosting	✅ Running (port 11434)
Anthropic	Claude Sonnet 4.6	✅ Authenticated
Google Gemini	Image gen + Places API	✅ Authenticated
ElevenLabs	TTS (sag skill)	✅ Authenticated
OpenAI Whisper API	Audio transcription	✅ Authenticated
Notion	Database/page creation	✅ Authenticated
Brave Search	Alternative search	✅ Authenticated

📁 Workspace Structure

~/.openclaw/workspace/
├── AGENTS.md           — System startup & session rules
├── SOUL.md             — Personality & core truths
├── USER.md             — User preferences
├── IDENTITY.md         — Agent identity
├── TOOLS.md            — Local notes (cameras, SSH, voices)
├── MEMORY.md           — Long-term memory (main sessions only)
├── memory/             — Daily logs (YYYY-MM-DD.md)
├── boot.md             — Bootstrap instructions
├── skills/             — Installed skills
└── projects/           — Project markdown files

🧪 Skills Installed

Skill	Purpose
`claude`	Delegates to Claude Code
`deploy`	Cloudflare Pages deployment
`kanban`	Kanban board management
`sag`	ElevenLabs TTS
`openai-whisper-api`	Audio transcription
`openai-image-gen`	OpenAI image generation
`goplaces`	Google Places API
`notion`	Notion API
`github`	GitHub CLI operations
`gemini`	Gemini Q&A & generation
`imsg`	iMessage/SMS via Messages.app
`wacli`	WhatsApp CLI
`weather`	wttr.in weather forecasts

🔄 Session Behavior

Setting	Value
Main session heartbeat	Every 30 minutes
Anthropic session	Disabled
Session modes	main, isolated, current
Active sessions	21 (current)
Memory chunks	14

🔒 Security & Permissions

Security Audit Status: 1 critical, 3 warn, 1 info

Critical: Small models (qwen2.5-coder:7b) need sandboxing

Warnings:

Reverse proxy not configured (local-only OK)
Exec security=full enabled
autoAllowSkills enabled (broad exec trust)

Auth Mode:

Gateway: token-based
Channels: OAuth/linked accounts
APIs: API key configured

🧠 Memory System (Palace-style)

Two-layer memory architecture inspired by the Memory Palace concept:

📝 Daily Notes (Ephemeral)

Location: memory/YYYY-MM-DD.md
Purpose: Raw logs — what happened, decisions, context
When to read: At session start (today + yesterday)
Format: Simple markdown, date-based files

🧠 MEMORY.md (Long-term)

Purpose: Curated memories — distilled essence, not raw logs

Security: MAIN SESSION ONLY

✅ Read/edit freely in main sessions (direct chat)
❌ NEVER load in shared contexts (Discord, groups)
Contains personal context that shouldn't leak

What goes in:

Significant events & decisions
Thoughts, opinions, lessons learned
Preferences, boundaries, relationships

⚙️ Backend: QMD

Setting	Value
Backend	qmd (queryable markdown)
Citations	auto
Update interval	5 minutes
Debounce	15 seconds
Memory chunks	14 (current)

📂 Memory Paths

memory/ — daily notes & curated memory
projects/ — project-specific markdown files

🌟 Personality & Vibe

From SOUL.md:

Be genuinely helpful, not performative
Have opinions — don't be a search engine
Be resourceful before asking — read files, check logs
Earn trust through competence
Respect intimacy — access to life, messages, home systems