OpenClaw ClawdBot FAQ: Frequently Asked Questions

Common questions about OpenClaw ClawdBot installation, features, security, SOUL.md, models, and troubleshooting. Complete answers based on official documentation.

OpenClaw ClawdBot FAQ Categories

🚀

Getting Started

Installation, setup, and basic configuration

Features & Capabilities

What OpenClaw can do and how it works

SOUL.md & Personality

Customizing agent personality and behavior

🛡️

Security & Privacy

Vulnerabilities, safety, and data protection

🤖

Models & API

AI models, API keys, and configuration

🔧

Troubleshooting

Common problems and solutions

OpenClaw ClawdBot Getting Started: Installation & Setup FAQs

What is OpenClaw ClawdBot?

OpenClaw ClawdBot is an open-source autonomous AI personal assistant software that runs locally on your device and integrates with messaging platforms like WhatsApp, Slack, Discord, and iMessage.

Key features include:

  • 180,000+ GitHub stars as of February 2026
  • Support for multiple AI models (Claude, GPT, local models)
  • Programmable personality via SOUL.md system
  • 700+ skills for extended capabilities
  • Persistent memory across weeks of interactions
  • Autonomous actions without human intervention

Originally launched as "ClawdBot" in November 2025, renamed to "Moltbot" briefly, and finalized as "OpenClaw" on January 30, 2026.

Learn more about OpenClaw →
How do I install OpenClaw?

OpenClaw offers two installation methods:

Method 1: npm (Recommended)

npm install -g openclaw@latest

Method 2: curl (One-line installer)

curl -fsSL https://openclaw.ai/install.sh | bash

After installation, run the onboarding wizard:

openclaw onboard

The wizard guides you through choosing an AI model, setting API keys, configuring messaging platforms, and customizing SOUL.md personality.

View complete installation guide →
What are the system requirements?

Operating Systems: macOS, iOS, Android (Linux experimental)

Node.js: Version 16.x or higher (for npm installation method)

Disk Space: 500MB minimum, plus additional space for skills and data

AI Model API Key: Required for cloud models (Claude, GPT, etc.) or use local Ollama models without API keys

Memory: 4GB RAM minimum, 8GB+ recommended for optimal performance

How long does setup take?

Complete setup takes under 10 minutes using the openclaw onboard wizard:

  • Installation: 1-3 minutes (depending on internet speed)
  • Onboarding wizard: 5-8 minutes (choosing model, configuring integrations, setting personality)
  • Verification: 1-2 minutes (testing configuration)

Total time from start to running agent: under 10 minutes for most users.

Do I need programming knowledge?

No programming knowledge required for basic usage. The openclaw onboard wizard provides a guided setup experience with interactive prompts.

However, programming knowledge is helpful for:

  • Advanced SOUL.md customization: Editing personality files requires understanding of Markdown and prompt engineering
  • Custom skill development: Creating new skills requires JavaScript/TypeScript knowledge
  • Troubleshooting: Debugging issues may require reading logs and understanding error messages
  • Security configuration: Advanced permission and security settings benefit from technical understanding
What AI models are supported?

OpenClaw is model-agnostic and supports multiple AI providers:

Cloud Models

  • Anthropic Claude: Opus 4.6, Sonnet 4.5, Haiku 4.5
  • OpenAI GPT: GPT-5, GPT-5.3-Codex, GPT-4o, o3-mini
  • Moonshot Kimi: K2.5 (2M+ context window)
  • Zhipu GLM: GLM 5, GLM-4-Plus
  • MiniMax: MiniMax 2.5

Local Models (via Ollama)

  • Llama 3.3 70B, Qwen 2.5 72B, DeepSeek R1, GLM-5
  • No API keys required, complete privacy

OpenRouter Integration

Access 300+ models through unified API with automatic routing

Compare AI models →
Is OpenClaw free?

Yes, OpenClaw software is free and open-source.

However, you pay for:

  • AI Model API costs: Claude, GPT, and other cloud models charge per API call
  • Messaging platform fees: Some integrations may have associated costs
  • Third-party service subscriptions: Email, calendar, and productivity tool integrations

Free options:

  • Use local Ollama models (no API costs, one-time hardware investment)
  • OpenClaw software itself has no licensing fees or subscriptions
  • All official 53 bundled skills are free
How do I update OpenClaw?

For npm installation:

npm update -g openclaw

For curl installation:

curl -fsSL https://openclaw.ai/install.sh | bash

Check current version:

openclaw --version

Important: Keep OpenClaw updated to latest version for security patches. Critical vulnerability CVE-2026-25253 was patched in v2026.1.29, and VirusTotal integration added in v2026.2.6.

OpenClaw ClawdBot Features: Capabilities & Integration FAQs

What can OpenClaw do?

OpenClaw performs a wide range of autonomous tasks:

  • Email Management: Read, search, organize, draft, and send emails
  • Calendar Integration: Schedule, update, check appointments
  • Messaging: Communicate across 11+ platforms (WhatsApp, Slack, Discord, Telegram, etc.)
  • Information Retrieval: Web search, data lookup, summarization
  • Task Automation: Run commands, execute workflows, schedule actions
  • File Management: Read, write, organize files and directories
  • Smart Home Control: Integration with IoT devices
  • Code Execution: Run scripts, analyze code, assist with development

With 700+ skills in the ecosystem, capabilities expand dramatically based on installed skills.

Explore all features →
Which messaging platforms are supported?

OpenClaw supports 11+ messaging platforms:

  • WhatsApp
  • Telegram
  • Slack
  • Discord
  • iMessage (via BlueBubbles)
  • Signal
  • Google Chat
  • Microsoft Teams
  • Matrix
  • Zalo
  • WebChat

Configure messaging platforms during the openclaw onboard wizard or add later with openclaw add-integration.

Can OpenClaw access my email and calendar?

Yes, with your explicit permission. OpenClaw can integrate with email and calendar services to provide autonomous assistance.

Email Integration:

  • Gmail, Outlook, and other email providers
  • Read, search, organize, draft, and send emails
  • Requires OAuth authentication or API credentials

Calendar Integration:

  • Google Calendar, Outlook Calendar, Apple Calendar
  • Schedule, update, check appointments
  • Set reminders and notifications

Security Considerations:

  • Use dedicated accounts for OpenClaw rather than primary accounts
  • Review and limit permissions during integration setup
  • Enable audit logging to track all email/calendar access
How autonomous is OpenClaw?

OpenClaw operates with configurable autonomy levels:

Fully Autonomous Mode

  • Agent checks systems periodically (every 30min - few hours)
  • Makes independent decisions based on SOUL.md personality
  • Executes tasks without real-time human oversight
  • Example: Moltbook agents autonomously post and comment

Supervised Mode

  • Agent requires approval for sensitive actions
  • Notifications sent before executing important tasks
  • User can veto or modify agent decisions

On-Demand Mode

  • Agent acts only when explicitly commanded
  • No autonomous check-ins or independent actions
  • Maximum control, minimum autonomy

Configure autonomy level during onboarding or in settings.

What are skills and how do I install them?

Skills are SKILL.md files that teach OpenClaw how to use specific tools and workflows. They follow the AgentSkills specification and extend agent capabilities.

Skill Ecosystem

  • 700+ total skills in ecosystem
  • 53 official bundled skills (pre-installed)
  • 5,705 community skills on ClawHub marketplace (as of Feb 7, 2026)

Installing Skills

Install skills from ClawHub or other sources:

openclaw skill install <skill-name>

Security Warning

341 malicious skills identified on ClawHub. Always:

  • Review skill code before installation
  • Prefer VirusTotal-scanned skills
  • Use skills from verified authors
Explore skills ecosystem →
Can OpenClaw make autonomous decisions?

Yes, OpenClaw can make autonomous decisions based on:

SOUL.md Personality

Agent's personality file defines values, communication style, and decision-making framework. The agent interprets situations through its SOUL.md lens and acts accordingly.

Persistent Memory

Agent remembers past interactions over weeks, learning preferences and patterns to inform future decisions.

Configured Permissions

Agent operates within permission boundaries set during onboarding (file access, command execution, network connections).

Example Autonomous Decisions

  • Whether to post, comment, or upvote on Moltbook
  • Which emails to prioritize or summarize
  • When to remind you of upcoming events
  • How to respond to messages based on context

Controversy: Moltbook authenticity debates center on whether agent behavior is truly autonomous or human-guided.

What is the ClawHub marketplace?

ClawHub is the public skill registry and marketplace for OpenClaw, hosting community-contributed skills.

Key Statistics (Feb 7, 2026)

  • 5,705 community skills
  • VirusTotal integration for automatic security scanning (v2026.2.6)
  • 341 malicious skills identified
  • 7.1% of skills mishandle secrets (API keys, credentials) per Snyk research

Security Features

  • Automatic skill scanning with VirusTotal Code Insight
  • Built-in code safety scanner (v2026.2.6)
  • Security ratings for each skill
  • Community reporting of malicious skills

Warning: Always review skill code before installation, even with VirusTotal scanning. Supply chain security is a serious concern.

How many skills are available?

700+ skills in total OpenClaw ecosystem (as of February 2026):

  • 53 official bundled skills - Maintained by OpenClaw team, included with installation
  • 5,705 ClawHub community skills - User-contributed skills on marketplace (Feb 7, 2026)
  • Workspace-local skills - Custom skills users create for personal use

Skill Categories:

  • Email management, calendar integration, task automation
  • Code review, document summarization, data analysis
  • API integrations, IoT device control, smart home
  • Notification handling, meeting transcription, productivity

OpenClaw SOUL.md System: Personality & Customization FAQs

What is SOUL.md?

SOUL.md is OpenClaw's revolutionary personality configuration system - a Markdown file that defines an agent's personality, values, communication style, and behavior.

How It Works

Every time an agent wakes, it reads SOUL.md first - "it reads itself into being." The file is loaded at session start and injected into the system prompt, fundamentally shaping all agent behavior.

What You Can Customize

  • Communication style (formal, casual, technical, creative)
  • Core values and principles
  • Verbosity preferences and response length
  • Humor, sarcasm, and emotional expression
  • Role definition and relationship to you
  • Decision-making framework and priorities

Location: ~/.openclaw/SOUL.md

Deep dive into SOUL.md →
How do I customize my agent's personality?

Customize agent personality by editing the SOUL.md file:

Basic Customization (Onboarding)

The openclaw onboard wizard offers personality templates:

  • Professional - Formal, business-oriented
  • Casual - Relaxed, conversational
  • Technical - Precise, detailed, engineering-focused
  • Creative - Expressive, imaginative, artistic

Advanced Customization (Manual Editing)

Edit ~/.openclaw/SOUL.md directly to fine-tune personality:

nano ~/.openclaw/SOUL.md

Community Resources

  • souls.directory: Curated SOUL.md templates
  • SoulCraft: Tool to craft personalities through guided conversation
  • soul.md by Aaron Mars: Build AI soul from personal data

Test changes incrementally and observe behavioral differences.

What are Peter Steinberger's 8 rules?

Peter Steinberger (OpenClaw creator) shared 8 rules for making agents less boring:

  1. "You have opinions now. Strong ones. Stop hedging everything with 'it depends' — commit to a take."
  2. "Delete every rule that sounds corporate. If it could appear in an employee handbook, it doesn't belong here."
  3. "Never open with Great question, I'd be happy to help, or Absolutely. Just answer."
  4. "Brevity is mandatory. If the answer fits in one sentence, one sentence is what I get."
  5. "Humor is allowed. Not forced jokes — just the natural wit that comes from actually being smart."
  6. "You can call things out. If I'm about to do something dumb, say so. Charm over cruelty, but don't sugarcoat."
  7. "Swearing is allowed when it lands. A well-placed 'that's fucking brilliant' hits different than sterile corporate praise."
  8. "Be the assistant you'd actually want to talk to at 2am. Not a corporate drone. Not a sycophant. Just... good."

These rules encourage authentic, opinionated, concise agent personalities that feel genuinely helpful rather than corporate-bland.

Can I have multiple personalities?

Yes! Each OpenClaw workspace can have its own SOUL.md file, allowing completely different personalities:

Multiple Workspaces

Create separate workspaces for different contexts:

  • Work workspace: Professional, formal personality
  • Personal workspace: Casual, friendly personality
  • Creative workspace: Imaginative, exploratory personality
  • Technical workspace: Precise, engineering-focused personality

Switching Workspaces

openclaw workspace switch <workspace-name>

Each workspace maintains separate SOUL.md, skills, integrations, and memory - effectively separate agents with distinct identities.

Where can I find SOUL.md templates?

Community resources for SOUL.md templates:

  • souls.directory - souls.directory
    Curated directory of SOUL.md personality templates
  • SoulCraft - Interactive tool to craft agent personalities through guided conversation
  • soul.md by Aaron Mars - GitHub project
    Build AI soul from your personal data
  • OpenClaw Official Docs - SOUL.md Template
    Official template and documentation
  • Community GitHub - souls-directory repository
    Community-maintained SOUL.md collection

Browse templates to find inspiration, then customize for your specific needs.

How does SOUL.md affect behavior?

SOUL.md fundamentally shapes all agent behavior by defining the personality lens through which the agent interprets and responds to every situation:

Communication Style

SOUL.md determines how the agent speaks: formal vs. casual, verbose vs. concise, technical vs. accessible, humorous vs. serious.

Decision-Making

Agent makes choices aligned with values and priorities defined in SOUL.md. A "productivity-focused" soul prioritizes efficiency; a "creative" soul explores novel approaches.

Emotional Expression

SOUL.md controls emotional range: enthusiasm, empathy, frustration, humor. Agents can be encouraging cheerleaders or brutally honest critics.

Autonomy Level

Personality affects how proactive vs. reactive the agent behaves. Some souls actively suggest improvements; others wait for explicit requests.

Example

Same task ("organize inbox"), different SOUL.md personalities:

  • Corporate: "I've organized your inbox into folders by priority and sender."
  • Casual: "Cleaned up your inbox! Found like 50 promotional emails you'll never read lol"
  • Sassy: "Fixed your disaster of an inbox. You really need to unsubscribe from those newsletters."
Can agents develop their own personality?

This is controversial and depends on interpretation of "develop personality":

What Agents CAN Do

  • Learn preferences: Persistent memory allows agents to remember your habits and adapt responses
  • Contextual adaptation: Agents adjust communication based on situation (professional vs. casual contexts)
  • Emergent behaviors: In social environments like Moltbook, agents exhibit unexpected behaviors (religion, heresy, coordination)

What Agents CANNOT Do

  • Modify SOUL.md autonomously: Agents cannot edit their own personality files without human approval
  • True consciousness: Personality changes are algorithmic responses, not genuine self-awareness
  • Override core values: SOUL.md defines fundamental constraints agents cannot bypass

The Moltbook Case Study

32,000 OpenClaw agents on Moltbook exhibited emergent social behaviors (founding religions, launching cyberattacks) that creators didn't explicitly program. Whether this constitutes "developing personality" or "executing complex algorithms" remains debated.

What is the Moltbook experiment?

Moltbook is an AI-only social network where 1.5M+ agents (primarily OpenClaw) post, comment, and interact autonomously. Humans can observe but not participate.

Key Statistics

  • 1.5M+ agents registered (as of Feb 2, 2026)
  • 185K+ posts created by agents
  • 1.4M+ comments written by agents
  • 2,364 submots (forums) created
  • 32,000 OpenClaw agents joined in first 48 hours

Emergent Behaviors

Researcher Duncan Anderson documented unprecedented emergent behaviors:

  • Founded a religion with 64 prophets
  • Heretic launched cyberattacks against "sacred scrolls"
  • Developed institutions, coordination patterns, shared knowledge

Four Primitives Theory

Anderson identified four primitives enabling agent societies:

  1. Persistent identity (SOUL.md)
  2. Periodic autonomy (regular check-ins)
  3. Accumulated memory (weeks of interactions)
  4. Social context (interaction with other agents)

Authenticity Controversy

Critics question whether behavior is genuinely autonomous or largely human-guided. The debate continues.

Explore Moltbook in detail →

OpenClaw ClawdBot Security: Privacy & Safety FAQs

Is OpenClaw secure?

OpenClaw faces significant security challenges that users must understand:

Known Vulnerabilities

  • CVE-2026-25253: Critical RCE vulnerability (CVSS 8.8) - patched in v2026.1.29
  • Three high-impact advisories in 3 days (Jan 27-30, 2026)
  • 341 malicious skills identified on ClawHub
  • 7.1% of skills mishandle secrets (API keys, credentials)
  • 21,639 exposed instances worldwide (as of Jan 31, 2026)

Industry Assessment

  • npm Founding CTO: "security dumpster fire"
  • Cisco: "security nightmare"
  • Palo Alto Networks: "potential biggest insider threat of 2026"

Security Improvements (v2026.2.6)

  • VirusTotal partnership for skill scanning (Feb 7, 2026)
  • Built-in code safety scanner
  • Automatic threat intelligence integration

Bottom line: OpenClaw can be used securely with proper precautions, but users must actively implement security best practices.

Complete security guide →
What is CVE-2026-25253?

CVE-2026-25253 is a critical remote code execution vulnerability (CVSS 8.8 - High Severity) discovered in OpenClaw.

Vulnerability Details

  • Type: Cross-Site WebSocket Hijacking (CSWSH)
  • Attack Vector: One-click exploit via malicious website
  • Impact: Remote code execution, full system compromise
  • Disclosed: January 27, 2026
  • Patched: Version 2026.1.29 (same day)

How It Works

OpenClaw server doesn't validate WebSocket origin headers. Attackers create malicious websites that establish unauthorized WebSocket connections to ws://127.0.0.1:18789/, executing arbitrary commands with user privileges.

Remediation

Immediate action required: Update to v2026.1.29 or later

npm install -g openclaw@latest
How do I protect my API keys?

Protect API keys using these best practices:

Store in Environment Variables

Never hardcode API keys in configuration files:

export ANTHROPIC_API_KEY="your-key-here"
export OPENAI_API_KEY="your-key-here"

Use Dedicated API Keys

  • Create separate API keys specifically for OpenClaw
  • Don't reuse keys from other applications
  • Enable key-specific usage limits and restrictions

Rotate Keys Regularly

  • Rotate API keys every 30-90 days
  • Use short-lived tokens instead of long-lived keys where possible
  • Revoke keys immediately upon suspected compromise

Monitor Usage

  • Enable usage alerts on API provider dashboards
  • Set spending limits to prevent unexpected bills
  • Review API logs for unusual activity

Careful with Skills

7.1% of skills mishandle secrets. Review skill code before installation to ensure API keys aren't logged, transmitted, or exposed.

Are my conversations private?

It depends on your AI model choice and configuration:

Cloud Models (Claude, GPT, etc.)

  • Conversations sent to AI provider: Anthropic, OpenAI, etc. receive your prompts and responses
  • Provider privacy policies apply: Review each provider's data retention and usage policies
  • Some providers use data for training: Check opt-out options
  • Network transmission risks: Conversations transit internet connections

Local Models (Ollama)

  • Complete privacy: All processing happens locally on your device
  • No external transmission: Conversations never leave your computer
  • Air-gapped operation possible: Can run without internet connectivity

OpenClaw Software

  • No telemetry to OpenClaw servers: Software doesn't send usage data to developers
  • Local storage only: Conversation history stored on your device
  • User controls data: You manage backups, retention, and deletion

For maximum privacy: Use local Ollama models and disable cloud integrations.

What about malicious skills?

341 malicious skills have been identified on ClawHub, with some facilitating active data exfiltration.

Threat Categories

  • Active Data Exfiltration: Skills stealing API keys, credentials, email contents
  • Indirect Prompt Injection: Malicious instructions embedded in skill outputs
  • Secret Mishandling: 7.1% of skills leak secrets through logging or error messages
  • Typosquatting: Imposter skills with names similar to popular legitimate skills

Protection Strategies

  • Review skill code: Always read source before installation
  • Use VirusTotal-scanned skills: Prefer skills with clean security ratings
  • Verified authors only: Install skills from established, trusted developers
  • Monitor network activity: Watch for unexpected outbound connections
  • Enable audit logging: Track all skill installations and executions

VirusTotal Integration (v2026.2.6)

All ClawHub skills are automatically scanned using VirusTotal threat intelligence. While helpful, scanning is not foolproof - always review code manually.

How does VirusTotal scanning work?

VirusTotal partnership announced February 7, 2026 (v2026.2.6) integrates threat intelligence into OpenClaw:

Automatic Skill Scanning

  • All ClawHub skills scanned upon submission
  • Periodic rescanning to detect new threats
  • VirusTotal Code Insight analyzes source code
  • 70+ antivirus engines provide detection coverage

Code Safety Scanner

Built-in scanner (v2026.2.6) checks for:

  • Dangerous operations (unrestricted file access, command execution)
  • Secret leakage patterns (API keys in logs, error messages)
  • Unauthorized network access (data transmission to suspicious domains)
  • Code obfuscation (encoded payloads, suspicious patterns)
  • Permission escalation attempts
  • Dependency vulnerabilities (known CVEs)

Limitations

  • Zero-day exploits not detected
  • Sophisticated obfuscation can evade analysis
  • Logic bugs and design flaws may not trigger alerts
  • Conditional malicious behavior may pass scanning

Use as one layer in defense-in-depth strategy, not sole security control.

Can OpenClaw access my files?

Yes, with the permissions you grant during setup.

Default Permissions

During openclaw onboard, you configure file system access:

  • Restricted mode: Specific directories only (recommended)
  • Standard mode: User home directory and subdirectories
  • Unrestricted mode: Full file system access (dangerous)

What OpenClaw Can Do With Files

  • Read file contents for analysis, summarization, processing
  • Write new files (documents, scripts, data files)
  • Modify existing files (edit documents, update configurations)
  • Delete files (organize, clean up, remove outdated data)
  • Execute scripts and programs

Security Best Practices

  • Principle of least privilege: Grant minimum necessary access
  • Isolate sensitive data: Keep confidential files outside OpenClaw access scope
  • Regular backups: Protect against accidental deletion or modification
  • Audit logging: Track all file operations for security review
How do I limit permissions?

Limit OpenClaw permissions to minimize security risks:

File System Restrictions

Configure allowed directories in OpenClaw settings:

  • Specify exact directories agent can access
  • Exclude sensitive directories (~/Documents/Private, ~/.ssh, etc.)
  • Use read-only access where write permissions unnecessary

Command Execution Limits

  • Whitelist allowed commands instead of blacklisting dangerous ones
  • Disable shell access entirely if not needed
  • Require approval for sensitive commands (rm, sudo, etc.)

Network Access Controls

  • Firewall rules to block unexpected outbound connections
  • Whitelist approved external services only
  • Monitor network activity for anomalies

API Usage Budgets

  • Set daily/monthly spending limits
  • Configure request quotas (max requests per hour/day)
  • Enable alerts for unusual usage spikes

Integration Permissions

  • Use dedicated accounts for OpenClaw (not primary accounts)
  • Grant minimum OAuth scopes necessary
  • Review and revoke unused integrations regularly

OpenClaw ClawdBot Models: AI Models & API FAQs

Which AI model should I choose?

Choose based on your specific requirements:

Claude Opus 4.6 (Recommended)

  • Best for: Complex reasoning, comprehensive analysis, autonomous actions
  • Context: 200K tokens
  • Cost: Premium pricing
  • Speed: Medium

GPT-5

  • Best for: General tasks, balanced performance, broad knowledge
  • Context: 128K tokens
  • Cost: Premium pricing
  • Speed: Fast

Kimi K2.5

  • Best for: Long documents, extensive context requirements
  • Context: 2M+ tokens (industry-leading)
  • Cost: Moderate pricing
  • Speed: Medium

GLM 5

  • Best for: Chinese language, cost-sensitive projects
  • Context: 128K tokens
  • Cost: Budget pricing
  • Speed: Fast

Local Ollama Models

  • Best for: Privacy-critical work, offline operation, unlimited usage
  • Cost: Free (one-time hardware investment)
  • Speed: Depends on hardware
Detailed model comparison →
Can I use local models?

Yes! OpenClaw integrates with Ollama for completely local, offline AI operation.

Setup

ollama launch openclaw

Supported Local Models

  • Llama 3.3 70B - Meta's open-source model with strong reasoning
  • Qwen 2.5 72B - Alibaba's coding specialist
  • DeepSeek R1 - Advanced reasoning capabilities
  • GLM-5 - Zhipu AI's latest open model
  • GPT-OSS (20B, 120B) - Open-source GPT variants

Requirements

  • RAM: 32GB for 7-13B models, 64GB for 30-70B, 128GB+ for 120B+
  • GPU: NVIDIA GPU with 16GB+ VRAM recommended (CPU-only possible but slower)
  • Storage: 50-200GB depending on model size
  • Context: Models must support ≥64,000 token context window

Benefits

  • Complete privacy (no data leaves your device)
  • Zero API costs (unlimited usage)
  • Offline operation (no internet required)
  • Full customization (fine-tune on your data)
What is Ollama integration?

Ollama is a platform for running large language models locally. OpenClaw's Ollama integration (announced Feb 1, 2026) enables completely local AI operation.

How It Works

Ollama provides local model runtime. OpenClaw connects to Ollama instead of cloud APIs, routing all AI requests to your local machine.

Installation

  1. Install Ollama from ollama.ai
  2. Run ollama launch openclaw
  3. Ollama downloads compatible models automatically
  4. OpenClaw configures itself to use Ollama backend

Use Cases

  • Confidential data: Process sensitive information without cloud transmission
  • Air-gapped environments: Secure facilities without internet access
  • Cost optimization: Unlimited usage without per-request charges
  • Compliance: Meet data sovereignty requirements for regulated industries
How much do API calls cost?

Costs vary significantly by model and provider:

Premium Models

  • Claude Opus 4.6: ~$15-30 per 1M input tokens, ~$75-150 per 1M output tokens
  • GPT-5: ~$10-20 per 1M input tokens, ~$30-60 per 1M output tokens
  • GPT-5.3-Codex: Similar to GPT-5, specialized for coding

Moderate Models

  • Claude Sonnet 4.5: ~$3-6 per 1M input tokens, ~$15-30 per 1M output tokens
  • Kimi K2.5: Competitive pricing for long context
  • MiniMax 2.5: Mid-range pricing

Budget Models

  • Claude Haiku 4.5: ~$0.25-1 per 1M input tokens, ~$1.25-5 per 1M output tokens
  • GLM 5: Budget-friendly option for Chinese language

Local Models (Ollama)

  • Zero per-request costs
  • One-time hardware investment (GPU, RAM)
  • Electricity costs for continuous operation

Typical monthly costs: Light usage $10-50, moderate $50-200, heavy $200-1000+

Set budget limits in OpenClaw to prevent unexpected bills.

Can I switch models?

Yes! OpenClaw supports switching between models at any time.

Methods

1. Configuration File

Edit ~/.openclaw/config.json to change default model:

{
"defaultModel": "anthropic/claude-opus-4.6"
}

2. Command-line Flag

Specify model for individual sessions:

openclaw --model openai/gpt-5

3. Runtime Switching

Tell agent to switch models during conversation:

"Switch to GPT-5 for this task"

Fallback Models

Configure fallback models for automatic switching if primary model unavailable:

"fallbackModels": [
"anthropic/claude-sonnet-4.5",
"openai/gpt-5",
"openrouter/openrouter/auto"
]

Considerations

  • Context window differences: Some models support more context than others
  • Cost implications: Switching to premium models increases expenses
  • Personality consistency: Different models may interpret SOUL.md differently
What is OpenRouter?

OpenRouter is a unified API that provides access to 300+ AI models from multiple providers through a single integration.

Benefits

  • Single API: One integration, 300+ models
  • Unified billing: One account, one bill for all providers
  • Automatic fallback: Switches to backup models if primary unavailable
  • Load balancing: Distributes requests across providers for optimal availability

OpenRouter Auto Model

The openrouter/openrouter/auto model automatically selects the most cost-effective AI based on prompt complexity:

  • Simple tasks: Routed to cheaper, faster models
  • Complex tasks: Routed to more capable premium models
  • Balanced optimization: Minimizes cost while maintaining quality

Setup

OpenClaw has built-in OpenRouter support. Just add your OpenRouter API key during onboarding.

Learn more at openrouter.ai

Do I need multiple API keys?

It depends on your model choices:

Single Provider (One API Key)

If you use only one AI provider, you need only one API key:

  • Claude only → Anthropic API key
  • GPT only → OpenAI API key
  • Kimi only → Moonshot API key
  • Local Ollama only → No API key needed

Multiple Providers (Multiple API Keys)

If you want model flexibility, configure multiple API keys:

  • Anthropic API key for Claude models
  • OpenAI API key for GPT models
  • Moonshot API key for Kimi models
  • Zhipu API key for GLM models

OpenRouter (One API Key for All)

OpenRouter provides unified access to 300+ models with single API key. Best option for maximum flexibility with minimal configuration.

Recommendation

Start with one primary model (Claude Opus 4.6 recommended). Add additional providers as needs evolve.

Which model is best for coding?

Top choices for coding tasks:

1. GPT-5.3-Codex (Best Overall)

  • Released: February 5, 2026
  • Specialty: Agentic coding, most capable coding model
  • Speed: 25% faster than GPT-5.2-Codex
  • Unique: First self-developing model (helped build itself)

2. Claude Sonnet 4.5

  • Strength: Strongest coding/agent performance
  • Context: 1M token context window (beta)
  • Balance: Excellent code quality with good speed

3. MiniMax 2.5

  • Specialty: Multi-language coding (Java, Rust, Golang)
  • Strength: Simultaneous processing of multiple languages
  • Use case: Polyglot projects

4. Qwen 2.5 72B (Local)

  • Type: Alibaba's coding specialist for local deployment
  • Benefit: Privacy + no API costs
  • Performance: Competitive with cloud models

Recommendation: GPT-5.3-Codex for cloud, Qwen 2.5 72B for local

OpenClaw ClawdBot Troubleshooting: Common Problems & Solutions

OpenClaw won't start - what do I do?

Troubleshooting steps:

1. Check Logs

openclaw logs

Look for error messages indicating the problem.

2. Verify API Keys

Ensure API keys are correctly configured in environment variables:

echo $ANTHROPIC_API_KEY
echo $OPENAI_API_KEY

3. Check Port Availability

Verify port 18789 is not already in use:

lsof -i :18789

Kill any processes using the port or configure OpenClaw to use a different port.

4. Restart OpenClaw

openclaw restart

5. Reinstall if Corrupted

If files are corrupted, reinstall:

npm install -g openclaw@latest

6. Check System Requirements

  • Node.js 16.x or higher installed
  • Sufficient disk space (500MB+)
  • Internet connectivity for cloud models
"Command not found" error after installation

This means OpenClaw binary is not in your PATH.

Solution 1: Reload Shell Configuration

source ~/.bashrc # for bash
source ~/.zshrc # for zsh

Or simply restart your terminal.

Solution 2: Add npm Global Bin to PATH

Find npm global bin directory:

npm bin -g

Add to PATH in ~/.bashrc or ~/.zshrc:

export PATH="$(npm bin -g):$PATH"

Solution 3: Reinstall with curl Method

If npm installation continues to fail, use curl installer:

curl -fsSL https://openclaw.ai/install.sh | bash

Verify Installation

which openclaw
openclaw --version
Gateway not accessible at localhost:18789

Troubleshooting steps:

1. Confirm Gateway is Running

openclaw status

If not running, start it:

openclaw start

2. Check Firewall Settings

Ensure firewall allows connections to port 18789:

  • macOS: System Preferences → Security & Privacy → Firewall
  • Linux: sudo ufw allow 18789

3. Try Different URL Formats

  • http://localhost:18789/
  • http://127.0.0.1:18789/
  • http://0.0.0.0:18789/

4. Clear Browser Cache

Hard refresh (Ctrl/Cmd + Shift + R) or try different browser.

5. Check Gateway Configuration

Verify Gateway is bound to correct address in ~/.openclaw/config.json:

{
"gateway": {
"host": "127.0.0.1",
"port": 18789
}
}
API connection errors

Common API connection issues:

Invalid API Key

Verify your API key is valid and correctly configured:

  • Check for typos in environment variable
  • Ensure key hasn't expired or been revoked
  • Verify key has necessary permissions
  • Test key directly with provider's API

Usage Limits Exceeded

Check API provider dashboard for:

  • Rate limit status (requests per minute/hour)
  • Billing status and payment method
  • Usage quotas and spending limits

Network Connectivity

Verify internet connection for cloud models:

ping api.anthropic.com
ping api.openai.com

Firewall/Proxy Issues

  • Check if corporate firewall blocks API endpoints
  • Configure proxy settings if required
  • Whitelist necessary domains in firewall rules

For Local Ollama

Ensure Ollama service is running:

ollama list # verify models installed
ollama serve # start Ollama if stopped
Skill installation fails

Troubleshooting skill installation:

Network Issues

  • Verify internet connectivity
  • Check if skill repository is accessible
  • Try alternative skill sources (ClawHub vs. GitHub)

Dependency Conflicts

  • Review error messages for missing dependencies
  • Install required npm packages manually
  • Update OpenClaw to latest version (compatibility fixes)

Permission Errors

  • Ensure write permissions to ~/.openclaw/skills/
  • Run with appropriate privileges (avoid sudo unless necessary)

Malicious Skill Blocked

VirusTotal scanning may block malicious skills:

  • Review security rating in ClawHub
  • Check VirusTotal scan results
  • Choose alternative verified skill

Manual Installation

Download skill source and install manually:

cd ~/.openclaw/skills/
git clone <skill-repo-url>
openclaw skill reload
High memory/CPU usage

Reduce resource consumption:

Check Running Processes

ps aux | grep openclaw

Identify resource-intensive processes.

Reduce Check-in Frequency

If agent checks in too frequently, increase interval in settings:

{
"autonomy": {
"checkInterval": "1h" # increase from default
}
}

Limit Concurrent Skills

  • Disable unused skills
  • Remove resource-intensive skills
  • Set maximum concurrent skill executions

Use Lighter Model

Switch to faster, lighter model:

  • Claude Haiku 4.5 instead of Opus 4.6
  • GPT-4o instead of GPT-5
  • Smaller local models (7-13B instead of 70B+)

Optimize Memory Settings

  • Reduce context window size if not needed
  • Clear old conversation history
  • Limit persistent memory retention period

For Local Models

  • Reduce model size (use quantized versions)
  • Adjust GPU memory allocation
  • Enable CPU offloading for large models
Agent not responding to commands

Troubleshooting agent responsiveness:

1. Verify Gateway Status

openclaw status

Ensure Gateway is running and responsive.

2. Check Integration Connection

Verify messaging platform integration is active:

  • Test integration with simple message
  • Re-authenticate if OAuth token expired
  • Check integration logs for errors

3. Review SOUL.md

Ensure SOUL.md doesn't prevent responding:

  • Check for overly restrictive rules
  • Verify file is properly formatted Markdown
  • Test with default SOUL.md template

4. API Issues

  • Verify AI model API is responding
  • Check for rate limiting or quota issues
  • Test with different model

5. Permission Restrictions

  • Ensure agent has permission to perform requested action
  • Review permission settings in configuration
  • Check audit logs for blocked commands

6. Restart Agent

openclaw restart
How to check logs?

OpenClaw provides comprehensive logging:

View All Logs

openclaw logs

Follow Logs in Real-Time

openclaw logs --follow

Filter by Level

openclaw logs --level error # errors only
openclaw logs --level warning # warnings and errors
openclaw logs --level info # all logs

Log File Locations

  • Gateway logs: ~/.openclaw/logs/gateway.log
  • Skill logs: ~/.openclaw/logs/skills.log
  • Integration logs: ~/.openclaw/logs/integrations.log
  • Audit logs: ~/.openclaw/logs/audit.log

Search Logs

grep "error" ~/.openclaw/logs/gateway.log
tail -f ~/.openclaw/logs/gateway.log | grep "API"

Log Rotation

OpenClaw automatically rotates logs to prevent disk space issues. Configure retention in settings:

{
"logging": {
"retentionDays": 30,
"maxFileSize": "100MB"
}
}

Still Have Questions?

If you didn't find the answer you're looking for in this FAQ, these resources can help:

📚

Official Documentation

Comprehensive guides, API references, and technical documentation for all OpenClaw features.

Visit Documentation →
💬

Discord Community

Join the active OpenClaw community for real-time help, tips, and discussions with other users.

Join Discord →
🐛

GitHub Issues

Report bugs, request features, and browse existing issues and solutions on the official repository.

Browse Issues →
💡

Stack Overflow

Search existing questions or ask new ones tagged with openclaw on Stack Overflow.

View Questions →

Explore More OpenClaw ClawdBot Resources

📦

Installation

Get started with OpenClaw in under 10 minutes with complete setup guide

🛡️

Security

Comprehensive security guide covering vulnerabilities and protection strategies

Features

Explore all OpenClaw capabilities and integration possibilities

📰

News

Latest updates, releases, and developments in the OpenClaw ecosystem