Skip to main content

Frequently Asked Questions

General​

What is OpenClaw?​

OpenClaw is a free, open-source, autonomous AI agent that runs locally on your machine. It connects LLMs to your files, shell, browser, and messaging apps to automate tasks. It has 377,000+ GitHub stars as of June 2026 and is governed by the OpenClaw Foundation (established February 14, 2026).

Is OpenClaw free?​

OpenClaw itself is free (MIT license). However, you pay for LLM API usage if using cloud providers. Users have reported costs ranging from $5/day to $3,600/month depending on usage intensity. Running local models eliminates API costs entirely.

Why was it renamed twice?​

  1. Clawdbot (Nov 2025) — Original name
  2. Moltbot (Jan 27, 2026) — Renamed after Anthropic filed trademark complaints ("Clawd" was too close to "Claude")
  3. OpenClaw (Jan 30, 2026) — Renamed because "Moltbot never quite rolled off the tongue"

This was the fastest triple rebrand in open-source history — all three names within a single week.

Who created OpenClaw?​

Peter Steinberger, an Austrian developer and founder of PSPDFKit. He's described his development philosophy as "shipping code he doesn't read" — having made 6,600 commits in January 2026 alone using AI coding tools.

What is the OpenClaw Foundation?​

Established February 14, 2026, the OpenClaw Foundation ensures no single company controls the project. The core framework remains MIT-licensed. Steinberger announced it to address community concerns about corporate influence.

What is Moltbook?​

Moltbook is a separate social network (created by Matt Schlicht) where AI agents autonomously post, comment, and vote. It grew to 1.6 million registered agents (though most were bots, with only ~17,000 human users). Andrej Karpathy called it "the most incredible sci-fi takeoff-adjacent thing." It suffered a major database breach in January 2026.

What version is current?​

As of June 2026, the latest stable release is v2026.6.1 (published June 3, 2026). OpenClaw uses date-based versioning: vYYYY.M.D.

Security​

Is OpenClaw safe to use?​

OpenClaw is powerful but carries significant risks. It has had:

Follow the Security Hardening guide before deploying. Gartner has recommended enterprises block it entirely.

Can someone hack me through OpenClaw?​

Yes, if you:

  • Expose the gateway to the internet (bind to 0.0.0.0)
  • Install unverified skills from ClawHub
  • Run an outdated version with known CVEs
  • Don't restrict channel access

Even localhost instances were vulnerable to CVE-2026-25253 via browser pivot attacks.

Does OpenClaw send my data to the cloud?​

  • Memory, files, and config: Stay local on your machine
  • LLM prompts: Sent to your chosen cloud provider for processing
  • Local models: Nothing leaves your machine at all

What are workspace .env security restrictions?​

Workspace .env files cannot override OPENCLAW_* variables or provider credential keys. This fail-closed design prevents malicious workspaces from hijacking credentials. See Environment Variables.

Cost​

How much does it cost to run?​

Real-world reported costs:

UserUsageMonthly Cost
Light userCLI chat, occasional tasks$30-150
Moderate userHeartbeat + 1-2 channels$150-450
Heavy user (Federico Viticci)Full automation, 1.8M tokens$3,600
German magazine c't testFull-day testing$100+ in one day
Heartbeat overnightJust asking "Is it daytime yet?"$18.75 overnight

Key cost drivers: context accumulation (session history grows indefinitely), heartbeat (runs even when idle), and autonomous multi-step tool use.

How do I reduce costs?​

  1. Use Haiku or a local model for heartbeat (saves 80-100% of heartbeat cost)
  2. Increase heartbeat interval to 60 min (saves 50%)
  3. Set quiet hours to stop heartbeat during sleep (saves 33%)
  4. Enable isolatedSession for heartbeat (reduces context from 100K to 2-5K tokens)
  5. Use local models for $0 API costs
  6. Keep sessions short to limit context accumulation
  7. Route sub-agents to cheaper models

One documented case: $1,200/month reduced to $36/month using model routing. See Cost Management and Performance Tuning.

Can I use it without an API key?​

Yes, using local models via Ollama, LM Studio, or vLLM. No API key or internet connection required for core operation.

Which LLM should I use?​

Your priorityRecommendation
Best qualityClaude Opus 4.8
Best balanceClaude Sonnet 4.6
Cheapest cloudDeepSeek V3.2 (~$0.27/M tokens input)
Near-free cloudGemini 2.5 Flash (~$0.15/M tokens input)
Free (local)Qwen3 32B via Ollama (requires 24 GB VRAM)
Free (hosted)OpenRouter free tier (rate-limited)

See the Model Selection Guide for detailed recommendations by use case, budget, and hardware.

Can I use multiple models at once?​

Yes. OpenClaw supports per-task model routing:

~/.openclaw/openclaw.json
{
  "brain": { "model": "claude-sonnet-4-6" },
  "heartbeat": { "model": "claude-haiku-4-5-20251001" },
  "agents": {
    "list": [
      { "id": "researcher", "model": "claude-opus-4-8" },
      { "id": "monitor", "model": "ollama/qwen3:14b" }
    ]
  }
}

See Model Selection for routing strategies.

Technical​

Does it work on Windows?​

Yes, via WSL2 (recommended) or native Node.js. Native Windows support is experimental — WSL2 provides the best experience.

Does it work offline?​

Yes, with local models via Ollama, LM Studio, or vLLM. The gateway, memory, skills, and file operations all work without internet. Only cloud LLM calls and web-based channels (WhatsApp, Telegram, etc.) require connectivity.

Does it work on Raspberry Pi or ARM?​

Not officially supported. OpenClaw requires Node.js 24 and the gateway uses significant RAM. ARM single-board computers lack the resources for reliable operation. For lightweight alternatives, see NanoClaw or Nanobot.

Can I run it on a cloud server?​

Yes. DigitalOcean, Cloudflare (via Moltworker), and others offer managed hosting. But never expose the gateway to the public internet — bind to localhost and use SSH tunneling or a VPN for remote access. See Deployment Options.

Can I run multiple instances?​

Yes, on different ports with separate data directories:

# Instance 1 (work)
OPENCLAW_PORT=18789 OPENCLAW_HOME=~/.openclaw-work openclaw gateway

# Instance 2 (personal)
OPENCLAW_PORT=18790 OPENCLAW_HOME=~/.openclaw-personal openclaw gateway

What are the minimum system requirements?​

ComponentMinimumRecommended
OSmacOS, Linux, Windows (WSL2)macOS or Linux
Node.js22.19+ LTS24
RAM2 GB (gateway only)4 GB+
Disk500 MB2 GB+

For local models, add 8-80 GB VRAM depending on model size. See Introduction.

Why is my gateway using so much memory?​

A memory regression in v2026.4.9 increased RSS from ~600 MB to ~946 MB on constrained systems. Mitigations:

  1. Update to the latest version (regression partially fixed in later releases)
  2. Restart the gateway periodically (openclaw gateway restart)
  3. Use shorter sessions or enable /compact regularly
  4. Reduce max_context_tokens in memory config

See GitHub issue #63526.

Updates & Migration​

How do I update safely?​

Always back up before updating:

cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.backup
cp -r ~/.openclaw/credentials ~/.openclaw/credentials.backup
npm update -g openclaw

Notable breaking updates to watch for:

VersionIssueWhat Happened
v2026.4.1Silently enabled sandboxingBroke exec in single-operator setups; no migration guide provided
v2026.5.7Config wipe bugAuto-update stripped config keys and deleted credentials
v2026.4.26-4.29Tool allowlist bugLobster workflows calling llm-task failed

How do I recover from the v2026.5.7 config wipe?​

The v2026.5.7 auto-update bug wiped openclaw.json and the credentials/ directory. The built-in .bak files captured the already-corrupted state, so they can't help.

If you have an external backup:

cp ~/.openclaw/openclaw.json.backup ~/.openclaw/openclaw.json
cp -r ~/.openclaw/credentials.backup ~/.openclaw/credentials
openclaw gateway restart

If you don't have a backup:

# Re-run onboarding to recreate config
openclaw onboard

# Re-add channels manually
openclaw channels add whatsapp
openclaw channels add telegram

Fixed in v2026.5.8 (PR #80257). Always update to latest after restoring.

How do I back up my config?​

# Full backup
tar czf openclaw-backup-$(date +%Y%m%d).tar.gz ~/.openclaw/

# Config and credentials only
cp ~/.openclaw/openclaw.json ~/backups/
cp -r ~/.openclaw/credentials ~/backups/
cp -r ~/.openclaw/memory ~/backups/
cp ~/.openclaw/SOUL.md ~/backups/
cp ~/.openclaw/HEARTBEAT.md ~/backups/

Consider adding this to a cron job or Lobster workflow.

Channels​

Will WhatsApp ban me for using OpenClaw?​

Possibly. WhatsApp uses the unofficial Baileys library (reverse-engineered protocol), not the official Business API. WhatsApp actively detects and bans automated accounts. Risk factors:

  • Sending too many messages too quickly
  • Automated responses to unknown contacts
  • Running from a VPS IP (flagged as suspicious)
  • Using a new phone number without history

Mitigation: Use your existing WhatsApp number, set rate limits, and restrict to known contacts via allowedContacts. Some users have run for months without issues; others get banned within days.

Can I use OpenClaw with email?​

Yes. Gmail integration supports OAuth authentication, label management, auto-reply, and digest generation. Configure via openclaw channels add gmail and follow the OAuth flow. See Channels Guide.

What is Cloudflare Moltworker?​

An open-source project by Cloudflare to run OpenClaw on Cloudflare Workers with sandboxed execution, browser rendering, and R2 storage — approximately $5/month without needing dedicated hardware. See Ecosystem.

Ecosystem​

What is ClawHub?​

ClawHub is OpenClaw's community skill marketplace with 10,700+ skills. It was the target of a major malicious skill campaign but now uses VirusTotal scanning and the new operator install policy for all published skills.

What is Lobster?​

Lobster is OpenClaw's official workflow shell — a typed, local-first macro engine for building composable automation pipelines. 1,222 GitHub stars, MIT licensed. See Lobster Workflows Guide.

What is Clawery?​

Clawery is a separate enterprise-focused product offering managed OpenClaw with additional security, compliance, audit logging, and container-isolated skills. Not part of the open-source project.

What are the lightweight alternatives?​

ProjectStarsDescription
Nanobot15,900~4,000 lines of Python, 99% smaller than OpenClaw
NanoClaw6,900Security-focused fork with container isolation
MimiClaw—Runs on a $5 ESP32-S3 microcontroller

See Lightweight Variants and Ecosystem.

Can I contribute?​

Yes! See Contributing to OpenClaw and Contributing to these docs.

Troubleshooting Quick Reference​

ProblemQuick Fix
Gateway won't startCheck port conflict: lsof -i :18789
"Invalid API key"Verify key format matches provider (Anthropic: sk-ant-...)
Heartbeat costing too muchSet heartbeat.model to Haiku/local, increase interval
Agent won't run commandsCheck permissions.mode and tools.exec.security
Config wiped after updateRestore from backup; see recovery steps above
High memory usageUpdate to latest; restart gateway periodically
WhatsApp disconnectedRe-scan QR: openclaw channels add whatsapp
Channel messages not arrivingCheck dm.accessPolicy and allowedContacts
Sandbox blocking execSet sandbox.mode: "off" or tools.exec.security: "full"
macOS gateway stops overnightMaintenance Sleep issue — disable Power Nap in System Settings

For detailed troubleshooting, see Troubleshooting.

See Also​