Frequently Asked Questions

Everything stays in Germany. We use Hetzner Cloud data centers in Falkenstein, Germany, so latency for European users is low and you're fully GDPR compliant. Your data never leaves the EU.

Under 60 seconds in most cases. We keep a pool of pre-warmed servers with all software already installed, so your instance is basically ready right away. You'll get an email once it's done, and you can watch the progress live in the dashboard.

Yes, any time you want. When you pause, we create a snapshot and temporarily remove the server. You won't be charged while it's paused. When you're ready to continue, we restore everything from the snapshot. One thing to keep in mind: if your balance stays at zero for 3 days, paused instances get deleted.

You can open a support ticket directly from the dashboard. We monitor all instances on our end too. If something critical goes down, we'll reach out to you before you even notice. We don't offer formal SLA guarantees, but we take uptime seriously.

You've got two options: the ClawHosters dashboard and direct SSH. The dashboard is probably the easiest way to manage things day to day. And SSH is there for the cases where you need to dig deeper. You can add your own public keys in the dashboard settings.

Yes, as many as you need. Each one is billed on its own. There's no hard limit on how many you can create, and you manage all of them from the same dashboard. If you're planning a larger deployment, reach out and we can talk about volume pricing.

We support Telegram, Discord, Slack, and WhatsApp. You set them up in the dashboard under "Configure Messenger." For Telegram, you'll need a bot token from BotFather. Discord and Slack work similarly. You grab a bot token from their developer portals. WhatsApp is a bit different since it uses a pairing flow through your instance's web interface. You can connect more than one platform to the same instance if you want.

If you see this status after provisioning, it just means the server is running but still needs an LLM API key. Head to "Configure AI" in the dashboard and add your key. Once that's done, the instance switches to "Running" on its own.

If you set a messenger channel's DM policy to "Pairing," new users need your approval before they can talk to your AI agent. You'll see pending requests in the dashboard and can approve or deny them one by one. If you'd rather skip that step, set the DM policy to "Open" and anyone can start a conversation right away.

Go to your instance in the dashboard and select "Rebuild" from the management options. A rebuild is a full factory reset. It wipes the container completely and deploys a fresh OpenClaw installation from scratch. Use it as a last resort when something is fundamentally broken and a simple restart doesn't help. You will need to go through the setup wizard again to reconfigure your LLM provider, messenger channels, and other settings. The process usually takes about a minute.

Yes. Go to the Access tab in your dashboard and enable SSH access. You can either use the provided key or add your own public key. Once enabled, you'll see the connection details (IP address and port) right there in the dashboard.

A rebuild wipes ALL data inside the container. This includes your LLM configuration, messenger tokens, chat history, cron jobs, custom files, SSH keys, and any other modifications you made. It is a complete factory reset.What is preserved: your server IP address, subdomain, instance tier, and billing mode. Everything else starts fresh. You will go through the setup wizard again to reconfigure your instance from scratch.

If your instance failed during restoration (e.g., the snapshot timed out or the server didn't come back up), you can try clicking "Resume" in the dashboard again. If that doesn't work, there may be an issue with the underlying snapshot. Contact support@clawhosters.com with your instance name and any error messages you see. We can manually trigger a restoration or provision a fresh instance with your saved configuration.

First, check that your instance is still running in the dashboard. If it shows "Running" but the bot is silent, go to the Channels tab and verify your bot token is still valid. Tokens can expire if you regenerate them in BotFather. If the token looks fine, try disconnecting and reconnecting the channel. That usually fixes it.

You just need a regular WhatsApp account on a dedicated phone number. No WhatsApp Business account is required. ClawHosters uses the WhatsApp Web protocol (Baileys) to connect. Go to the Channels tab in your dashboard and select WhatsApp. Your instance will show a QR code that you scan with your regular WhatsApp app to pair the connection. After pairing, messages flow through automatically.

Yes. You can connect Telegram, WhatsApp, Discord, and Slack all to the same instance at the same time. Every channel shares the same AI brain and configuration, so your assistant behaves consistently no matter where people reach it.

This is a known UI delay. The WhatsApp connection can take a few seconds to sync with the dashboard status. Wait about 30 seconds and refresh the page. If the status still shows "Waiting" after a minute, try disconnecting and re-pairing through the QR code flow. In rare cases, the session dropped silently. Re-pairing fixes it.

WhatsApp allows up to 4 linked devices per phone number, but if you reach that limit, adding a new linked device will disconnect the oldest one. If you have multiple WhatsApp Web sessions open, one of them may be your ClawHosters session. Check your linked devices in WhatsApp (Settings > Linked Devices) and make sure the ClawHosters session is still active. Also check that your phone has a stable internet connection, since WhatsApp requires the phone to stay online for the bridge to work. If disconnections keep happening, go to Settings and try "Doctor Fix" in the Troubleshooting section to reconnect channels.

Go to the Discord Developer Portal (discord.com/developers) and create a new application. Under the "Bot" tab, create a bot and copy its token. Make sure to enable "Message Content Intent" under Privileged Gateway Intents, otherwise the bot won't be able to read messages. Then in your ClawHosters dashboard, go to the Channels tab, select Discord, and paste the bot token. Finally, invite the bot to your Discord server using the OAuth2 URL generator in the Developer Portal (select the "bot" scope and the permissions your bot needs).

Yes. You can connect Slack to your OpenClaw instance just like Telegram, WhatsApp, and Discord. Set it up in the Channels tab in your dashboard. You'll need to create a Slack app in the Slack API portal, add the bot token, and invite the bot to your workspace. All four platforms can run simultaneously on the same instance.

Error 4014 means the "Message Content Intent" is not enabled for your Discord bot. Go to the Discord Developer Portal (discord.com/developers), select your bot application, navigate to the "Bot" tab, and enable "Message Content Intent" under Privileged Gateway Intents. Without this, your bot can't read message contents and will silently ignore everything. After enabling it, reconnect the Discord channel in your ClawHosters dashboard.

Start with "Doctor Fix" in Settings under the Troubleshooting section. It reconnects channels and fixes stale plugin state without disrupting your instance. If that doesn't work, try "Restart Gateway" (also non-disruptive, reloads the connection layer). For persistent issues, disconnect the channel in the dashboard, wait a few seconds, and reconnect it. If you're on WhatsApp, you'll need to re-pair through the QR code. For Telegram, just re-enter the bot token.

Open your instance in the dashboard and go to the Settings tab. You can change things like the system prompt, model selection, temperature, and other parameters there. Changes apply within seconds through live reload, so you don't need to restart anything.

Skills are add-on capabilities for your AI agent, like web search, image generation, or file handling. You can browse and install them from ClawHub directly in your dashboard. If you need something custom, you can also build your own skills and upload them to your instance.

Config changes apply through live reload, so a bad value can cause immediate problems. Common mistakes: invalid API key format, unsupported model names, or typos in configuration keys. Go to Settings and try "Repair Config" in the Troubleshooting section first. It scans for common problems and fixes them automatically. If that doesn't help, check the Logs tab for the specific error. You can also use "Doctor Fix" to resolve stale state without disrupting anything.

The Config Editor in your dashboard validates keys as you type. If you enter an invalid key, it shows a warning. Stick to the keys shown in the editor's autocomplete. For advanced configuration, refer to the OpenClaw documentation. Making changes through the dashboard Config Editor is always safer than editing configuration files directly, since it catches errors before they're applied.

These are two different authentication paths to the same provider: openai/ uses an API key from platform.openai.com. You pay per token and get the rate limits of your API tier. openai-codex/ uses your ChatGPT subscription (Plus, Pro, Team) via OAuth. You use your subscription quota with its own rate limits. This matters for model configuration: If you signed in via Codex OAuth, only openai-codex/ models work (e.g., openai-codex/gpt-5.1). Models with the openai/ prefix (e.g., openai/gpt-5.2) require a separate API key and will fail if none is set.Common pitfall: You configure openai/gpt-5.2 as a fallback but only have OAuth. The fallback fails, and OpenClaw falls back to your primary model, which might already be rate-limited. Fix: Use openai-codex/ models or clawhosters/ models (the managed models included in your plan) as fallbacks instead.

Fallback models take over when your primary model is unavailable (e.g., due to rate limits or outages). Without configured fallbacks, the agent just fails instead of switching to a backup.You can configure fallbacks through the dashboard in the Settings tab, or via command line (SSH): text Copy openclaw models fallbacks add clawhosters/deepseek-reasoner openclaw models fallbacks add clawhosters/deepseek-chat Your plan includes free managed models that work well as fallbacks: clawhosters/deepseek-reasoner (DeepSeek R1, good for reasoning) clawhosters/deepseek-chat (DeepSeek V3, fast) clawhosters/gemini-2.5-flash-lite (Gemini Flash Lite, fast and cheap) Also check your agent settings and cron jobs. If they reference models from providers you haven't configured (e.g., nvidia/ without an nvidia API key), those will fail and unnecessarily load your primary model through fallback cascades.

OpenClaw has a built-in cron scheduler that lets your assistant run tasks on a schedule or at a specific time. You can ask your assistant to "check the news every morning" (recurring) or "remind me at 3pm tomorrow" (one-time). The assistant creates cron jobs automatically when you make these requests.There are two schedule types: "every" runs a job on a repeating interval (e.g., every hour, every day at 9am), and "at" runs a job once at a specific time and deletes itself afterward.Each cron job also has a payload type that determines what happens when it fires: agentTurn: The assistant runs a full conversation turn and can send you a message. Use this for reminders and anything that should produce a visible response. systemEvent: Injects a note into the session context silently. This does not send you a message or notification. Use this for background context updates. For reminders that actually reach your chat, the job must use agentTurn with announce delivery mode. If your assistant creates systemEvent jobs for reminders instead, install the remind-me skill from ClawHub to teach it the correct approach.You can view and manage your cron jobs in the Cron tab of your dashboard. The run history shows whether each job executed successfully and whether a message was delivered.

Each LLM provider uses a specific format string. For Anthropic (Claude models), use anthropic-messages (e.g., anthropic-messages/claude-sonnet-4-20250514). For OpenAI, use openai-completions (e.g., openai-completions/gpt-4o). For Google Gemini, use google (e.g., google/gemini-2.5-flash). For OpenRouter, use openrouter (e.g., openrouter/anthropic/claude-sonnet-4). The format matters. Using the wrong one will cause API errors. Check the Addons tab in your dashboard for the full list of supported providers and their format strings.

Not all models handle tool calling equally well. Tool calling is how your assistant interacts with OpenClaw's built-in tools (like creating cron jobs, sending messages, or browsing the web). Lightweight and "fast" model variants often struggle with complex tool schemas, which means they might fail silently or create malformed tool calls.Models that work well for tool use and scheduling: xAI Grok 4.1 Fast: Excellent tool calling, available globally. Z.AI GLM-5: Strong tool calling, good alternative if xAI isn't available in your region. Anthropic Claude Sonnet 4: Very reliable (requires API access from supported regions). OpenAI GPT-4o: Solid tool use (requires API access from supported regions). Models that tend to struggle with tool calling: Gemini Flash Lite and similar ultra-lightweight models. Non-reasoning "fast" variants of smaller models. Some DeepSeek models when accessed through managed proxies (timeout issues). If your reminders aren't working, the model might be creating the wrong type of cron job (systemEvent instead of agentTurn). Install the cron-mastery skill from ClawHub to teach your assistant the correct approach. Then test by asking for a simple reminder and checking the Cron tab to see if a job was created with the right type.

Some LLM providers restrict direct API access from certain regions. As of 2025, both Anthropic (Claude) and OpenAI (GPT) block API requests from several countries and territories, including Hong Kong, mainland China, Russia, and others. If you're in an affected region, API key validation will fail even with a valid key.Alternatives that work globally or have fewer restrictions: xAI (Grok): Available in most regions. Grok 4.1 Fast is excellent for tool calling and scheduling. Z.AI / Zhipu: The Z.AI Coding Plan addon gives you access to GLM-5 (strong reasoning and tool use), GLM-4.7, and GLM-4.5 Air. Available globally. Google Gemini: Available in most countries. Gemini 2.5 Flash is fast and capable. DeepSeek: Available globally. Good for general conversation, though some models can be slow for tool-heavy tasks. All of these can be configured as your primary model or as fallbacks in the Settings tab. If you're unsure which providers work in your region, try adding the API key in your dashboard. The key validation check will tell you immediately if the connection works.

Model naming conventions vary by provider, but these labels generally mean:Fast: Optimized for speed and lower cost. Responses come back quickly, but the model may be less accurate on complex tasks. Good for simple conversations and quick replies.Reasoning: The model thinks step-by-step before answering. This makes it better at complex tasks like creating cron jobs with the right parameters, multi-step tool use, and following detailed instructions. The trade-off is slower response times and higher token usage.Non-reasoning: Skips the internal reasoning step entirely. Fastest response times, but can miss nuances in tool schemas or complex instructions. Fine for casual chat, but less reliable for scheduling and automation.For reminders, cron jobs, and anything involving tool calling, reasoning variants or standard (non-suffixed) models tend to work more reliably than non-reasoning fast variants. If your assistant keeps failing to create cron jobs or call tools correctly, switching from a "fast" or "non-reasoning" variant to the standard or "reasoning" version of the same model often fixes it.

OpenClaw has a built-in memory system that indexes your workspace files and makes them searchable during conversations. The key pieces: MEMORY.md: A curated file in your workspace root with your long-term context (identity, preferences, key info). This file gets special treatment: always indexed first, never affected by temporal decay. Writing a solid MEMORY.md is the single highest-impact thing you can do for memory quality. memory/*.md files: Daily logs, project notes, workflow docs. Automatically indexed and searchable. memorySearch: The search engine that retrieves relevant context. Supports hybrid search combining BM25 keyword matching with vector semantic search. To enable memorySearch with hybrid mode, go to the Settings tab and update your config: json Copy "agents": { "defaults": { "memorySearch": { "enabled": true, "provider": "openai", "model": "text-embedding-3-small", "query": { "hybrid": { "enabled": true, "vectorWeight": 0.7, "textWeight": 0.3 } } } } } If you're using OpenRouter for embeddings, set remote.baseUrl to https://openrouter.ai/api/v1 and add your API key under remote.apiKey.You can also enable temporal decay (recent memories rank higher, 30-day half-life) and MMR diversity (avoids redundant results from similar daily notes) under query.hybrid. Check the OpenClaw documentation for the full memory configuration reference.

Yes, with some caveats depending on your tier.QMD (Query Markdown Database) is an alternative memory backend that runs locally inside your instance. It downloads three small GGUF models (~2GB total) and does BM25 + vector + LLM re-ranking entirely on-device, with no external API calls for search. To set it up: Enable SSH access in your dashboard SSH in and run: bun install -g https://github.com/tobi/qmd Add "memory": { "backend": "qmd" } to your config QMD uses about 600MB of extra RAM. On Balanced and Pro tiers that's no problem. On Budget tier it could get tight depending on what else your agent is doing. The model download happens once on first run.Graphiti uses a Neo4j graph database for temporal knowledge graphs with auto-recall and auto-capture. It's a powerful concept, but it requires running a full Neo4j instance alongside your agent. That's not feasible on ClawHosters currently, since each instance runs a single Docker composition and adding extra database services isn't supported.Obsidian pairs well with your instance if you have the Obsidian skill installed. You can use Obsidian on your computer to curate and organize your memory files, then sync them to your instance's workspace using a sync plugin. This gives you a nice visual editing experience locally while your agent accesses the same files on the server.For most setups, the built-in memorySearch with hybrid mode gives excellent results without extra dependencies. See the FAQ entry above for setup instructions.

OpenClaw does not have maxContextMessages or maxContextTokens settings. Adding unknown keys to your config will cause validation errors and can crash the gateway. Only use keys that appear in the Config Editor's autocomplete.The right way to control context size is through three built-in mechanisms that work together: Context Pruning (agents.defaults.contextPruning): Trims old tool results before each LLM call. The default mode is cache-ttl. Key settings include keepLastAssistants (how many recent assistant turns to preserve fully, default 3) and softTrim.maxChars (shrinks oversized tool outputs, default 4000 chars). This is the most direct lever for reducing per-call token usage. Compaction (agents.defaults.compaction.mode: "safeguard"): Automatically summarizes conversation history when the context approaches the model's token limit. Instead of crashing or truncating, OpenClaw condenses older turns into a summary and keeps going. This is already active by default. Session Reset (session.reset.mode: "daily"): Clears the entire conversation history once a day at a configured hour. Already enabled by default on ClawHosters instances. Keeps long-running conversations from accumulating unbounded context. For cost control specifically, model choice matters more than context tuning. The clawhosters/ models included in your plan (like clawhosters/deepseek-chat or clawhosters/gemini-2.5-flash-lite) are free to use and work well for routine tasks. Setting up fallback models so cheaper models handle simple requests, keeping conversations shorter through daily session resets, and using context pruning to trim bloated tool outputs all reduce costs more effectively than trying to hard-cap context size.Message-count-based pruning (like "only send the last N messages") is a commonly requested feature that doesn't exist in OpenClaw yet. It may come in a future version. For now, the combination of context pruning, compaction, and session reset covers most use cases well.

Yes. Your instance's configuration (openclaw.json) is automatically backed up twice a day at 05:00 and 17:00 UTC. This is a free feature included with every instance, no add-on needed. The backups capture your LLM settings, channel tokens, personality, skills, cron jobs, and all other config values.Retention works in tiers: all backups from the last 7 days are kept, plus one backup at the 14-day mark and one at the 31-day mark. Anything older than 37 days is deleted. Manual backups are never automatically removed.You can view, preview, and restore config backups from Settings > Config Backups in your dashboard.

Go to Settings > Config Backups, find the backup you want, and click the restore icon. You'll need to type your instance name to confirm, since restoring overwrites your current config and restarts the instance. The restore runs in the background and takes under a minute.Before restoring, you can click Preview to see the backup contents with API keys masked. If you've rotated API keys since the backup was created, you'll need to update them after the restore.It's a good idea to create a manual backup of your current config before restoring an older one. That way you can go back if needed.

Config backups only save the configuration file (openclaw.json). They're lightweight (a few KB), free, and run automatically. Use them to recover from broken config changes.The Backup Add-on creates full server snapshots through Hetzner. These capture everything: chat history, uploaded files, installed packages, custom modifications, and the config. It's a paid add-on and runs once per day.If you just changed a setting and broke something, a config backup gets you back in seconds. If you need to recover data or files, you need the Backup Add-on.

Yes. Go to Settings > Config Backups and click Backup Now. You can add a description like "Before switching to GPT-5" to help identify it later. Manual backups are kept indefinitely (they're never auto-deleted), so they serve as reliable save points.

Every plan comes with free AI included: Gemini 2.5 Flash Lite, DeepSeek R1, DeepSeek V3, and Kimi K2.5, with daily and monthly token limits. No API key needed to get started. If you want a different model (like Claude or GPT-4), more capacity, or unlimited usage, you can switch to BYOK and plug in your own API key from any supported provider (Anthropic, OpenAI, Google, DeepSeek, Mistral, Groq, or OpenRouter). BYOK is free. You just pay your provider directly.

That depends on how important your data is. The Backup add-on automatically creates snapshots of your instance with 7, 30, or 90 day retention. All backups are stored on Hetzner infrastructure. On top of that, a manual snapshot is created whenever you pause your instance. So even without the add-on, you have a basic safety net.

Yes. You can add or remove add-ons at any time from the dashboard. Billing adjusts proportionally, so you only pay for what you actually use.

Nothing. BYOK is free. You pay your API provider directly (OpenAI, Anthropic, whoever you choose) and that's it. Your keys are stored encrypted and never logged. If you already have API access or need a specific provider, BYOK is probably the way to go.

Open your instance and go to the AI Setup tab. You can switch between the included free AI models (Gemini 2.5 Flash Lite, DeepSeek R1, DeepSeek V3, and Kimi K2.5) and BYOK mode where you add your own API key from any supported provider. The change takes effect within seconds. For providers not in the dropdown, click "Add Custom Provider" in the AI Setup tab to add any OpenAI-compatible provider directly.

Yes. In the AI Setup tab, click "Add Custom Provider" to add any provider with an OpenAI-compatible API. Pick from presets (Featherless AI, Together AI, Fireworks AI, Ollama) or choose "Manual setup" for any other provider. Enter the base URL, your API key, and at least one model ID. After saving, you can set it as your primary model right away.For advanced configuration (custom cost values, reasoning flags, etc.), you can also edit the provider block directly in Settings > Config Editor under models.providers.

Yes. Your instance can have multiple LLM providers configured in parallel. For example, Claude (Anthropic) as the main model and Ollama on your own server as an additional model. You can request a specific model in conversations anytime, e.g. "Use ollama/llama3.1:8b for this task". If your local model is unreachable (e.g. server is off), your agent automatically falls back to the next available model. This gives you privacy for sensitive tasks and full power for everything else. Setting up a local LLM over ZeroTier is covered in the "Use Your Own LLM at Home" guide.

If you want your agent to use your local model by default (e.g. for GDPR compliance), reach out to support. We can set your default model to your local one, e.g. ollama/llama3.1:8b. Your agent then primarily uses the local model and only falls back to cloud models when the local one is unreachable. Prerequisite: ZeroTier must be set up and your LLM server needs to be running. Details in the ZeroTier LLM guide.

No. The Storage add-on is a separate disk that gets mounted alongside your server, not an extension of the built-in one. They serve different purposes: Built-in disk (40-160 GB depending on tier): Contains the operating system, Docker, OpenClaw itself, and all installed skill dependencies. You can't move these system files elsewhere. Storage add-on (your extra volume): A separate disk for your own files. Documents, media, downloads, project data. Inside your agent, it's available at /app/data/customer-storage/. If your built-in disk is getting full, it's usually from skill dependencies (Python packages, AI models, etc.) rather than your own files. You can tell your agent to save files to /app/data/customer-storage/ to keep the built-in disk lean. Check the "Do skills use disk space?" FAQ below for more details on what takes up space.

Yes, and sometimes a lot. When a skill has system dependencies (Python packages, CLI tools, media libraries), those get installed permanently inside the container the first time the skill runs. A single skill with heavy dependencies can add 500 MB to 2 GB of disk usage.This is especially noticeable on the Budget tier (40 GB) where system overhead already takes up a significant portion. If you have several skills with system dependencies (like video processing, web search, or document parsing skills), they can collectively use 3-6 GB.Uninstalling a skill removes its config but does not remove the system packages it installed. To fully reclaim that space, you'd need to rebuild the instance (which wipes everything and starts fresh).To check your skill disk impact, look at the Dependencies section in the Skills tab. Skills with many listed dependencies tend to use more disk.

Most skill issues come down to missing dependencies. Many skills need an API key or a specific add-on to function. For example, skills that use web search need a search API key configured. Check the skill's description in ClawHub for its requirements. If everything looks correct, try "Doctor Fix" in Settings, under Troubleshooting, to reset stale plugin state.

No. If you use a monthly subscription (which most users do), Claws are optional. You only need them if you choose daily billing or if you exceed your free managed LLM usage limits. Monthly subscriptions cover everything for standard usage.

Claws are our internal currency. You buy a package, and we deduct Claws daily for your active services. We've got five packages: €10 (650 Claws), €20 (1,380 Claws, 6% bonus), €40 (2,900 Claws, 12% bonus), €80 (6,100 Claws, 17% bonus), and €160 (13,000 Claws, 25% bonus). The bigger the package, the more bonus Claws you get.

We accept all major credit and debit cards through Stripe: Visa, Mastercard, American Express, Discover, and more. Apple Pay and Google Pay are also supported. You can also pay with cryptocurrency via Coinbase Commerce (Bitcoin, Ethereum, and other popular coins).

Because your instance is provisioned right away, we can't offer refunds. By purchasing, you waive the 14-day EU withdrawal right. If something goes wrong on our end (like a service outage we caused), reach out and we'll look into issuing a credit.

Your instance gets paused automatically. We'll warn you at 3, 2, and 1 day of remaining balance so you've got time to top up. Once paused, you have 3 days to add more Claws. After that, the instance is permanently deleted. Don't worry though, we create a snapshot before pausing so your data isn't just gone.

Every day at 01:00 Berlin time (CET/CEST), we deduct Claws from your balance based on your tier. Budget costs 60 Claws/day, Balanced costs 105 Claws/day, and Pro costs 175 Claws/day. If you create an instance mid-day, your first day is prorated based on the hours left until the next billing cycle. Add-ons are billed separately. You can check your usage anytime in the dashboard.

There's a one-time setup cost in Claws when creating a new instance: 150 Claws for Budget, 300 for Balanced, 450 for Pro. This covers the cost of provisioning your server. Monthly subscriptions don't have a setup fee.

Claws are consumed in two ways. Daily billing users have a fixed daily deduction based on their tier (60/105/175 Claws per day for Budget/Balanced/Pro), plus any active add-on costs. Monthly subscribers only consume Claws if they exceed their free managed LLM usage limits. In that case, extra LLM usage is charged per token from your Claws balance. You can track all usage in the dashboard.

It depends on your billing mode. Daily billing users: Your instance is paused entirely. We create a snapshot of your server, then the server is deleted to stop charges. You have 3 days to top up. After that, the snapshot is permanently deleted. Once you add more Claws, the instance is automatically restored from the snapshot. Monthly subscription users: If you run out of Claws used for managed LLM extra usage, the extra AI features stop working until you top up. Your instance itself keeps running normally since it is covered by your subscription.

Your bot doesn't stop when you hit 100% of your free managed LLM tokens for the day. Instead, the system switches to deducting from your Claws balance for any additional AI usage. If you have Claws available, everything keeps running seamlessly. If you don't have Claws, the extra AI features stop until the next day when your free tokens reset. Your instance itself stays online either way.

A few things help. Long conversations accumulate context, which means every new message sends the entire history to the AI. Use /new regularly to start fresh conversations. Disable skills you're not actively using, since each skill adds overhead to every request. Check if you have cron jobs running frequently, as those consume tokens too. And if you're on a model with a large context window, consider switching to one with a smaller window if you don't need the extra capacity.

For monthly subscriptions, you can upgrade directly from your instance settings in the dashboard. Go to your instance, open settings, and select a higher tier. The upgrade takes 2-3 minutes - your instance will be briefly offline while the server is replaced with one matching the new tier. Your data and configuration are preserved throughout. Daily-billed instances cannot change tiers. To switch tiers on daily billing, create a new instance on the desired tier and delete the old one.

No. The server infrastructure doesn't support switching from a larger tier to a smaller one. If you want to move to a lower tier, you'd need to cancel your current instance and create a new one on the desired tier. This means setting up your bot configuration again (LLM key, Telegram bot, personality, etc.). Your old instance data won't carry over automatically.

If a coupon was applied to your account and you want to remove it, contact support@clawhosters.com. Coupons are tied to billing records and can't be removed through the dashboard. Include your account email and which coupon you'd like removed, and we'll handle it.

All transfers are encrypted with TLS/SSL, and stored data uses AES-256 encryption. The servers sit in German Tier-3+ data centers with physical access controls. We run regular security audits, and everyone on the team who handles data is trained accordingly.

Yes. All data is stored in Germany and never leaves the EU. You have full GDPR rights, including access, rectification, erasure, and data portability. If you need to reach our data protection contact, send an email to support@clawhosters.com.

Only you. Our support team can see server metadata when troubleshooting, but they can't touch your application data unless you give explicit permission. Third parties like Hetzner, Stripe, and Coinbase only process the data they need to run their part of the service.

Every instance runs in its own isolated Docker container with strict resource limits. Firewall rules block unwanted traffic by default, and brute-force protection secures SSH access. If you need SSH access, it's available and secured with key-based authentication.

Your data gets deleted within 30 days after cancellation. You can request a full data export before that happens. We're required to keep billing and payment records for 10 years under German tax law. Server logs are automatically wiped after 90 days.

You can sign in with email and password, or use Google Sign-In. We recommend using a strong, unique password for your ClawHosters account.

Contact our support team at support@clawhosters.com and request account deletion. Your data will be removed within 30 days. Billing and payment records are kept for 10 years as required by German tax law, but all personal and instance data is fully erased.

No. ClawHosters instances are built for OpenClaw hosting. Additional ports (e.g. 80, 443 for custom web servers or proxies) are not opened for security reasons. Every instance is protected by multiple firewall layers that only allow the ports OpenClaw needs. This architecture protects all customers and cannot be customized per instance. If you want to run a web server or proxy alongside OpenClaw, we recommend using a separate VPS for that.

Reach out to support with both your current and new email address. We need to verify your identity before making the change to keep your account secure.

Start by checking the Logs tab in your dashboard for error messages. That usually tells you exactly what's going wrong. If you spot something obvious, fix it in your config. If not, go to Settings and open the Troubleshooting section. Try these steps in order: Repair Config scans your configuration for common problems and fixes them automatically. Doctor Fix runs a diagnostic inside the container and fixes stale plugin state. This is non-disruptive, so your sessions stay active. Restart Gateway reloads the gateway process without restarting the container. Also non-disruptive, takes about 15 seconds. Restart Container is the last resort. It kills active sessions and takes about 60 seconds. If none of these work, open a support ticket and include the log output from the Logs tab.

On the login page, click "Forgot Password" and enter your email. You'll get a reset link that's valid for 24 hours. If the email doesn't show up or something else goes wrong, reach out to support@clawhosters.com. We'll need to verify your identity before resetting it manually.

This happens when your Claws balance runs out. The system pauses the instance automatically to prevent unexpected costs. To get back up, top up your balance and click "Resume" in the dashboard. Your data is safe as a snapshot for 3 days. After that, the instance gets permanently deleted, so don't wait too long.

Most instances are ready in under 60 seconds. If yours is taking longer, it's likely a temporary capacity issue on our end. Give it a few minutes and refresh the status page in your dashboard. The system monitors stuck provisioning jobs automatically and will retry if something went wrong.

The most common reason is an instance or add-on you forgot was still running. Go to your dashboard and check what's currently active. Daily billing happens at 01:00 Berlin time (CET/CEST) for every active service. You can review the full transaction history to see exactly what was charged and when. If something still looks off, contact support and we'll sort it out.

A few things can cause this. The most common one is cron jobs blocking the message queue. If you have scheduled tasks running, they can hold up message processing. Check the Logs tab for processing times to see where the delay is.Long conversations also slow things down. As context accumulates over many messages, each new response takes longer because the model processes the full history. Use the /new command to start a fresh conversation and see if that helps.Beyond that, check your LLM provider's status page for outages. And keep in mind that model choice matters. Cheaper models aren't always faster. Gemini Flash and Deepseek tend to respond quickly, while larger models naturally take more time.

Tweak your system instructions to be more specific about what you expect. Lowering the temperature makes responses more focused and less creative. If the answers are still off, try a stronger model. Also check whether your context window might be too small for the conversation length you're working with.

Config changes apply through a live reload. If a bad config crashes the instance, go to Settings and open the Troubleshooting section. Use "Repair Config" to scan and fix common config problems automatically. If that doesn't help, check the Logs tab for the specific error message.Common causes are invalid API key format, an unsupported model name, or malformed JSON in advanced settings. You can also try "Doctor Fix," which resolves stale plugin state without disrupting the instance.

The "Reset to Defaults" function in the dashboard resets your entire configuration to the initial state. All custom settings (model selection, skills, environment variables, system prompt, cron jobs, messenger config) are lost. This action cannot be undone.If you had a backup before the reset: go to the Backups tab in your dashboard and restore the backup that was created before the reset. Check the timestamps carefully. Backups created after the reset contain the empty state.If you didn't have a backup: you'll need to reconfigure your settings manually. Go to Settings and set up your model, skills, and messenger connections again.Tip for the future: enable the Backup add-on before making major changes to your configuration. That way you always have a restore point. You can find the Backup add-on in the Backups tab of your dashboard.

The dashboard shows "Unhealthy" when the health check can't reach your instance. This usually means the container is stuck or overloaded. Go to Settings and open the Troubleshooting section. Start with "Doctor Fix" (non-disruptive). If that doesn't help, try "Restart Gateway." As a last resort, use "Restart Container" (this kills active sessions).If the instance stays unhealthy after all of these, open a support ticket and include the log output from the Logs tab.

You'll find these under Settings in the Troubleshooting section: Repair Config: Scans your configuration for common problems and fixes them automatically. May briefly restart the instance. Doctor Fix: Runs a diagnostic check inside the container. Fixes stale plugins, reconnects channels. Non-disruptive, your sessions stay active. Restart Gateway: Reloads the gateway process without restarting the container. Non-disruptive, takes about 15 seconds. Good first step for connection issues. Restart Container: Full container restart. This is disruptive and kills active sessions. Use only as a last resort. Takes about 60 seconds. Automatically runs Doctor Fix afterward. Clear Session Locks: Removes stale lock files. Use when your instance seems frozen but the status shows "Running." Approve Device: Approves CLI pairing requests from your terminal.

This means your LLM provider is rejecting requests because you've exceeded their rate limits. It's not a ClawHosters issue. It's a limit set by your API provider (Anthropic, OpenAI, etc.) based on your account tier.Both Anthropic and OpenAI use a tier system where new accounts start with very restrictive limits. Upgrading your tier is usually the fix.Anthropic (Claude) tiers Anthropic has 4 usage tiers. Tier upgrades happen automatically once you purchase enough credits at console.anthropic.com. Tier 1 ($5 in credits): 50 requests/min, 30,000 input tokens/min Tier 2 ($40 in credits): 1,000 requests/min, 450,000 input tokens/min Tier 3 ($200 in credits): 2,000 requests/min, 800,000 input tokens/min Tier 4 ($400 in credits): 4,000 requests/min, 2,000,000 input tokens/min The jump from Tier 1 to Tier 2 is 20x more capacity. If you're hitting rate limits on Claude, upgrading to Tier 2 for $40 in total credits almost always solves it.OpenAI (GPT) tiers OpenAI has 5 usage tiers. Unlike Anthropic, OpenAI also requires a waiting period (days since your first payment) before you can advance. Tier 1 ($5 paid): 500 requests/min, 30,000 tokens/min (GPT-4o) Tier 2 ($50 paid + 7 days): 5,000 requests/min, 450,000 tokens/min Tier 3 ($100 paid + 7 days): 5,000 requests/min, 800,000 tokens/min Tier 4 ($250 paid + 14 days): 10,000 requests/min, 2,000,000 tokens/min Tier 5 ($1,000 paid + 30 days): 10,000 requests/min, 30,000,000 tokens/min OpenAI's Tier 1 allows more requests per minute (500 vs Anthropic's 50), but the token-per-minute limit is similarly tight. The jump to Tier 2 ($50 + 7 day wait) gives you 10x more capacity. Manage your tier at platform.openai.com under Settings.OpenAI Codex OAuth (ChatGPT subscription) If you signed in with OpenAI Codex OAuth (your ChatGPT Plus/Pro subscription instead of an API key), different rate limits apply. OAuth tokens have lower limits in external tools like OpenClaw than direct API keys. This means Codex may still work fine in the app while OpenClaw is already hitting the limit.The reason: OpenAI distinguishes between usage inside their own apps (Codex CLI, ChatGPT) and usage via OAuth in third-party tools. Third-party limits are stricter.If you're using OAuth and hitting rate limits, you have three options: Configure fallback models: Use the included managed models (e.g., clawhosters/deepseek-reasoner) as fallbacks so your agent keeps working when OAuth hits the limit. Switch to an API key: An OpenAI API key (platform.openai.com) has significantly higher limits than OAuth. Set it up as an openai provider instead of openai-codex. Reduce usage: Fewer concurrent agents, cron jobs, and background tasks means less token consumption per minute. Important: Models with the openai/ prefix require an API key. Models with openai-codex/ use OAuth. If you only have OAuth, openai/ models won't work as fallbacks because there's no API key. Use clawhosters/ models as fallbacks instead.Other providers Google (Gemini), DeepSeek, and OpenRouter also have rate limits, but they tend to be more generous at entry level. Check your provider's dashboard for specifics.What to do right now Check your tier in your provider's console (console.anthropic.com or platform.openai.com) Upgrade your tier by purchasing more credits if you're on Tier 1 Add a fallback model in your Settings tab. Your plan includes free models (Gemini 2.5 Flash Lite, DeepSeek R1, DeepSeek V3) that take over automatically when your primary model hits rate limits. Without a fallback, the agent just fails instead of switching to a backup.

This error comes from the ClawHub registry, not from ClawHosters. ClawHub limits how many install requests can come from the same IP address in a short time. Since your instance runs on a VPS, it may share an IP range with other users, which makes it easier to hit the limit.Wait about 5 minutes and try installing the skill again from the Skills tab. The rate limit resets after a short cooldown. If it keeps failing after a few retries, open a support ticket and we'll install the skill manually for you.Note: the "Skill found" checkmark in the install dialog only confirms that the URL format is valid. It doesn't check whether the skill is actually downloadable at that moment. So you can see "Skill found" and still hit a rate limit during the actual install.

Some ClawHub skills need extra setup after installation. Check the skill's page on clawhub.ai for instructions. Common things to look for: Environment variables: Some skills need API keys or config values. Go to the Skills tab, find the skill, and check if it lists required environment variables. You can set them right there in the dashboard. Dependencies: Some skills require system packages (like a CLI tool or Python library). The Skills tab shows a "Dependencies" section with install buttons for each one. Click the install button and wait for it to finish. Restart needed: After installing a skill with dependencies, try "Restart Gateway" from Settings to pick up the changes. If the skill is installed and enabled but still doesn't respond, check the Logs tab for error messages mentioning the skill name.

Several things share your instance's disk, and most of them are system-level, not your data. Here's the typical breakdown for a Budget tier (40 GB): OpenClaw Docker image (~11 GB). The base application with all its dependencies. Skill dependencies (varies, can be 2-6 GB). Skills that need system packages (Python libraries, CLI tools, media processors) install them permanently inside the container. The more skills you add, the more this grows. System reserve (5 GB). A reserved file that acts as emergency buffer if the disk fills up. Playwright browsers (~650 MB). Pre-installed for web browsing and automation skills. Your data (sessions, memory, config). Usually small, often under 100 MB. Update backup images. After an OpenClaw update, a backup of the previous version may be kept temporarily. To check what's using space, open the File Explorer tab in your dashboard. It shows your files and overall disk usage. If you have SSH access enabled, run du -h --max-depth=1 / on the server for a directory-level breakdown.To free up space: Remove skills you don't use. Uninstalling a skill won't reclaim its already-installed system packages, but it prevents future growth. A rebuild gives you a clean slate (you'll need to reconfigure everything). Clean up old sessions. Run openclaw sessions cleanup to prune old conversation data. Contact support. We can check for leftover backup images or other reclaimable space on your instance. If disk stays above 80% regularly, consider upgrading to the Balanced tier (80 GB) for more headroom. See Understanding Resource Limits for tier details.

When you ask your assistant to remind you about something at a specific time, it creates a cron job (a scheduled task). For the reminder to actually send you a message, the cron job needs to use the right configuration. Specifically, it needs an "agentTurn" payload with "announce" delivery mode. If your assistant creates a "systemEvent" job instead, the reminder runs silently in the background without ever sending you a notification.Whether your assistant sets this up correctly depends on the AI model you're using. Lightweight models (like Gemini Flash Lite) may not know how to create the right type of cron job for reminders. More capable models handle this better.The easiest fix: install the cron-mastery skill from ClawHub. Go to the Skills tab in your dashboard and search for "cron-mastery". This skill teaches your assistant exactly how OpenClaw's scheduling system works, including how to create reminder jobs that actually deliver messages to your chat at the scheduled time. It covers cron expressions, timezone handling, and the difference between silent background jobs and jobs that send you a notification.Important: Some community-made reminder skills on ClawHub (like "remind-me-pro") require Python tools (uv or pip) that aren't available on ClawHosters. If you install a reminder skill and see errors about missing tools in the Logs tab, uninstall it and stick with cron-mastery instead. It works without any extra dependencies.Reminders "forgotten" overnight? If your assistant says it will remember something but the reminder is gone the next day, the likely cause is the daily session reset. Session resets clear conversation context (at 4 AM by default). For a reminder to survive across sessions, it needs to be saved as an actual cron job, not just noted in the conversation. With cron-mastery installed, your assistant knows to create a persistent cron job instead.After installing the skill, try setting a reminder again. If it still doesn't work, check the Cron tab in your dashboard to see if jobs are being created, and look at the Logs tab for any errors.

Some ClawHub skills are built with Python scripts and require tools like uv (a Python package runner) or pip (a Python package installer) to function. ClawHosters containers don't include Python, uv, or pip, so these skills can't run on your instance. You'll typically see errors like "pip: not found" or "uv: not found" in the Logs tab.This isn't a bug. It's a compatibility issue between the skill's requirements and the ClawHosters environment. Not every skill on ClawHub is designed to work in every setup.What to do: Check the Logs tab for the specific error message. If it mentions a missing binary (uv, pip, python3, etc.), the skill needs Python tools. Uninstall the skill from the Skills tab. It won't work without the required tools. Look for an alternative on ClawHub that doesn't require Python. Skills that are pure SKILL.md files (knowledge-only skills) always work. Skills that use the exec tool with Python scripts may not. Check the skill's ClawHub page before installing. If it lists uv or pip under requirements or dependencies, it won't work on ClawHosters. For reminders specifically, the cron-mastery skill is a knowledge-only skill that works without any extra dependencies. See "My scheduled reminders aren't being sent" above.

Error code 1000 is a normal WebSocket close code. It means the connection was closed cleanly, not that something crashed. You'll see this during config reloads, after tool call completions, or when the gateway reconnects after a routine health check.If it appears occasionally in your logs, it's harmless and you can ignore it. It's part of normal gateway operation.If you're seeing it constantly and messages stop going through, that's a different situation. Go to Settings and open the Troubleshooting section. Try "Restart Gateway" first (non-disruptive, takes about 15 seconds). If the problem continues, try "Restart Container" as a last resort.

This happens when your conversation gets too long for the model's context window. Every message, tool result, and workspace file adds to the context. When it exceeds the limit, the model can't process tool calls anymore, including file writes. You'll see the agent output text in chat just fine (shorter responses), but writing a file requires the model to handle the full context plus the file content plus the tool call structure. That combination pushes it over.The error in your Logs tab looks like: Context overflow: prompt too large for the model. Try /reset (or /new) to start a fresh session, or use a larger-context model.Quick fix Send /new as a standalone message to your bot. This clears the conversation history and gives you a clean context window. Then ask the agent to write the file again in that fresh session.Prevent it from happening Enable auto-compaction. This automatically summarizes older conversation history to keep the context manageable. Ask your agent: "Enable compaction mode safeguard." Or set it manually in the Config Editor: agents.defaults.compaction.mode to safeguard. Use a model with a larger context window. DeepSeek V3 Chat has 131K tokens. Gemini 2.5 Flash Lite (included in your plan) has 1 million tokens. For tasks that involve long conversations and file writing, switching to Gemini Flash Lite as your primary model helps a lot. You can change this in the AI Setup tab. Keep conversations shorter for file tasks. If you need the agent to write a report, start a fresh session first. Don't ask for the file write at the end of a long back-and-forth. Why fallback models don't always help If your primary model overflows, the agent falls through your fallback chain. But fallback models often have even smaller context windows (especially free models). So the overflow just happens again on the next model in line.

If you're using OpenRouter and see 404 No endpoints available matching your guardrail restrictions and data policy, your OpenRouter privacy settings are blocking free model endpoints.OpenRouter requires you to opt in to logging and training data for free models. Without that, the free tier won't serve any requests.How to fix it Go to openrouter.ai/settings/privacy Make sure logging is enabled for free models Check that you haven't blocked too many providers in your provider settings If you have "Zero Data Retention" (ZDR) enabled, some free endpoints won't be available. Disable ZDR or switch to paid models. After updating the settings, your openrouter/free and openrouter/auto fallback models should start working again.

Open the support ticket system in your dashboard and select the "Bug Report" category. Include the steps to reproduce the issue, what you expected to happen, and what actually happened. Screenshots or error messages help us fix things faster.

You can read the full Terms of Service at clawhosters.com/terms. They cover things like our Acceptable Use Policy, how billing works, liability limits, and how we handle disputes. When you use ClawHosters, you're agreeing to those terms.

The short version: don't do anything illegal or harmful. That means no malware, no spam, no phishing, no copyright violations, no crypto mining without permission, no DDoS attacks, no network scanning, and no hosting of illegal content. If you break these rules, your account gets terminated immediately with no refund.

We provide our services "as is" and can't guarantee 100% uptime at all times. Our liability is capped at whatever you've paid us in the last three months. We're not responsible for data loss, outages caused by third parties, force majeure events, or any indirect damages.

German law applies. The UN Convention on Contracts for the International Sale of Goods is excluded. If there's ever a dispute, we'll try to sort it out directly first. The EU also provides an online dispute resolution platform at ec.europa.eu/consumers/odr/.

Because we start provisioning your server right away, you waive the 14-day right of withdrawal under EU consumer law when you purchase. This is shown clearly at checkout and you'll need to confirm it before completing your order.

ClawHosters is run by Daniel Samer in Germany, as a service of Yixn.io. You can reach us at support@clawhosters.com. The legal notice (Impressum) is at yixn.io/en/imprint. For anything related to data protection, just use the same contact address.

ClawHosters is managed hosting for OpenClaw AI agents. We take care of the server infrastructure, security, and maintenance so you can focus on your AI agent instead. All servers run in German data centers and are fully GDPR compliant.

Once you've registered and logged in, hit "New Instance" in your dashboard. Pick your billing mode (daily or monthly), choose a tier (Budget, Balanced, or Pro), and confirm. Your instance is usually ready in under 60 seconds. You'll get an email as soon as it's up.

There are three tiers to choose from. Budget (EUR 19/month or 60 Claws/day) gives you 2 vCPU, 4 GB RAM, and 40 GB storage. Balanced (EUR 35/month or 105 Claws/day) comes with 4 vCPU, 8 GB RAM, and 80 GB storage, and it's the most popular option. Then there's Pro (EUR 59/month or 175 Claws/day) with 8 vCPU, 16 GB RAM, and 160 GB storage for heavier workloads. All tiers include 20 TB traffic.

Not at all. The setup walks you through everything with a step-by-step wizard.

Yes. For monthly subscriptions, you can upgrade your tier directly from your instance settings in the dashboard. The upgrade takes 2-3 minutes and your data and configuration are preserved. Daily-billed instances cannot change tiers - to switch, create a new instance on the desired tier.

You can email us at support@clawhosters.com. Balanced and Pro customers get priority support. There's also an option to create a support ticket right from your dashboard. We typically reply within 24 hours on business days.

Yes. New customers get a free trial period with full access to all features. Your card isn't charged until the trial ends, and you can cancel anytime before that.

No. Every plan includes free AI models (Gemini 2.5 Flash Lite, DeepSeek R1, DeepSeek V3, and Kimi K2.5) with daily and monthly token limits. No API key needed to get started. If you want a different model (like Claude or GPT-4), more capacity, or unlimited usage, you can bring your own API key (BYOK) from any supported provider. That's free too, you just pay your provider directly.

An API key is a credential from an AI provider (like Anthropic, OpenAI, or Google) that lets your OpenClaw instance use their AI models. Think of it as a password that connects your assistant to a specific AI brain. You don't need one to get started though. Every plan includes free AI models (Gemini 2.5 Flash Lite, DeepSeek R1, DeepSeek V3, and Kimi K2.5) with daily and monthly token limits. API keys are only relevant if you want to use a different model or need unlimited usage through BYOK (Bring Your Own Key).

ClawHosters supports all the big AI providers. Anthropic (Claude models, recommended for coding and automation), OpenAI (GPT-4o), Google (Gemini), OpenRouter (access to 200+ models through one key), DeepSeek, Mistral, and Groq. You can also use OpenRouter for access to privacy-focused and open-source models.

Every provider has their own process, but it works pretty much the same everywhere. Sign up on their website, add a payment method, and create an API key. For Anthropic, go to console.anthropic.com. For OpenAI, go to platform.openai.com. For Google, go to aistudio.google.com. For OpenRouter, go to openrouter.ai. If you have a Claude Pro or Max subscription, you can also run claude setup-token in your terminal to generate a token.