LLM Add-on (BYOK vs Managed)

How LLM Works on ClawHosters

Every OpenClaw instance can use a large language model for conversations, tasks, and automations. ClawHosters gives you two ways to connect one: bring your own API key or use a managed pack.

BYOK (Bring Your Own Key)

BYOK lets you use API keys you already have from providers like Anthropic, OpenAI, or Google. ClawHosters does not charge anything for BYOK. You pay your provider directly.

Supported Providers

• Anthropic: Claude Opus, Sonnet, Haiku
• OpenAI: GPT-4, GPT-4 Turbo, GPT-3.5, o1
• Google AI: Gemini models
• DeepSeek: DeepSeek V3 and others
• OpenRouter: Access to multiple providers through a single key
• Mistral: Mistral models
• Groq: Fast inference models

Setting Up BYOK

1. Open your instance in the ClawHosters dashboard
2. Go to the AI Setup tab
3. Under Your API Keys (BYOK), click Add Provider
4. Choose your provider from the dropdown
5. Paste your API key
6. Click Save

The system runs a quick test call against your provider to confirm the key works. If verification succeeds, your instance is ready to use that provider's models.

Info: API keys are encrypted at rest. ClawHosters never stores them in plain text.

Why Choose BYOK

• No extra cost from ClawHosters
• Use any model your provider supports, including the latest releases
• No token limits on the ClawHosters side (your provider's limits apply)
• Good for teams that already have API agreements with a provider

Managed Packs

If you do not have your own API keys or want a simpler setup, managed packs give you a fixed monthly token allowance at a set price. ClawHosters handles the API keys and routing.

Available Tiers

Tier	Model	Best For
Eco	DeepSeek V3	General tasks, best value per token
Standard	Gemini Flash-Lite	Fast responses, reliable performance
Premium	Claude Haiku	Complex reasoning, nuanced conversations

Each tier uses a specific model. You cannot swap models within a tier.

Pricing and Token Allowances

Eco (DeepSeek V3)

Pack	Tokens/Month	Monthly Price
Starter	1,000,000	€3
Standard	5,000,000	€10
Pro	15,000,000	€25

Standard (Gemini Flash-Lite)

Pack	Tokens/Month	Monthly Price
Starter	1,000,000	€5
Standard	5,000,000	€15
Pro	15,000,000	€40

Premium (Claude Haiku)

Pack	Tokens/Month	Monthly Price
Starter	1,000,000	€12
Standard	3,000,000	€30
Pro	8,000,000	€70

Setting Up a Managed Pack

1. Open your instance in the ClawHosters dashboard
2. Go to the AI Setup tab
3. Under Included AI, select Managed Pack
4. Choose a tier (Eco, Standard, or Premium)
5. Choose a pack size (Starter, Standard, or Pro)
6. Confirm your subscription

The LLM is available immediately after subscribing. Token usage starts counting from the first request.

Token Tracking

Every LLM request is tracked in real time. Both input tokens (what you send) and output tokens (what the model responds) count toward your pack limit.

You can check your current usage on the AI Setup tab in the dashboard:

• Tokens used: How many tokens you have consumed this period
• Tokens remaining: How many you have left in your pack
• Usage percentage: A visual indicator of consumption

What Happens When You Run Out

If your managed pack runs out of tokens, further LLM requests return an error until:

• Your pack resets at the start of the next billing period
• You upgrade to a larger pack

Your instance keeps running normally. Only LLM requests are affected. All other features (messaging, channels, automations) continue working.

Info: BYOK subscriptions do not have token limits on the ClawHosters side. Usage is tracked for your reference but never blocked.

BYOK vs Managed: Quick Comparison

	BYOK	Managed Pack
Cost	Free (you pay your provider)	€3–70/month
Model choice	Any model your key supports	Fixed model per tier
Token limits	None from ClawHosters	Pack-based (1M–15M)
Setup	Add your API key	Subscribe to a pack
Best for	Power users, specific model needs	Simplicity, predictable costs

Model Privacy and Jurisdiction

When your instance sends a prompt to an LLM, that request leaves your German VPS and goes to the model provider's servers. Where those servers are and how the provider handles your data depends on the model.

Included Models (Free with Every Plan)

These models are available on every ClawHosters plan at no extra cost. ClawHosters pays for the API access.

Model	Provider	HQ / Jurisdiction	Subject to US CLOUD Act	Data Training	Servers in China
Gemini 2.5 Flash Lite	Google	USA	Yes	No (paid API usage excluded per Google's API terms)	No
DeepSeek R1	DeepSeek	China (Hangzhou)	No	Policy less transparent, check current API terms	Yes
DeepSeek V3 (Chat)	DeepSeek	China (Hangzhou)	No	Policy less transparent, check current API terms	Yes
Kimi K2.5	Moonshot AI	China (Beijing)	No	Policy less transparent, check current API terms	Yes

Google's API terms state that paid API data is not used for model training. Since ClawHosters pays for Gemini API access, your usage counts as paid and falls under this exclusion.

DeepSeek and Moonshot AI have less transparent data handling policies. Their API terms can change. If this matters to you, verify their current terms directly.

Managed Pack Models

Model	Provider	HQ / Jurisdiction	Subject to US CLOUD Act	Servers in China
DeepSeek V3 (Eco tier)	DeepSeek	China (Hangzhou)	No	Yes
Gemini Flash-Lite (Standard tier)	Google	USA	Yes	No
Claude Haiku (Premium tier)	Anthropic	USA	Yes	No

Anthropic states that API data is not used for training by default.

BYOK Provider Jurisdictions

Provider	HQ	Subject to US CLOUD Act	EU-Based
Anthropic (Claude)	USA	Yes	No
OpenAI (GPT)	USA	Yes	No
Google AI (Gemini)	USA	Yes	No
DeepSeek	China	No	No
Mistral	France	No	Yes
OpenRouter	USA	Yes	No
Groq	USA	Yes	No

EU-Compliant Alternative

If EU jurisdiction matters to you, Mistral is the strongest option among supported BYOK providers. Mistral is a French company with servers in the EU, fully GDPR-compliant, and not subject to the US CLOUD Act or Chinese data laws. You can get an API key at mistral.ai and add it under AI Setup > Your API Keys (BYOK). BYOK is free on our end.

What ClawHosters Controls

Your instance infrastructure runs on Hetzner Cloud in Germany. Chat history, configuration, and all stored data stay in Germany. ClawHosters routes LLM requests through our infrastructure but does not store prompt content or responses. The distinction is: your data at rest stays in Germany, but inference happens at the model provider's servers.

For full details on data handling, see Data Handling and Privacy.

Provider Failover

For managed packs, if the primary provider experiences downtime, the system automatically routes requests through a backup provider. This happens transparently. You do not need to do anything. Response quality stays the same because the backup uses an equivalent model.

BYOK requests use your key directly and do not have automatic failover.

Troubleshooting

"No LLM configured" error

• Verify you have an active LLM subscription or BYOK credential for this instance
• Check the AI Setup tab to confirm the subscription status is "Active"

BYOK key verification failed

• Double-check that you copied the full API key without extra spaces
• Make sure the key has not been revoked or expired at your provider
• Confirm the key has the correct permissions (most providers require at least chat/completions access)

"Usage limit exceeded" on managed pack

• Your pack's token allowance for this period is exhausted
• Upgrade to a larger pack or wait for the next billing cycle
• Consider switching to BYOK if you regularly exceed your limit

Slow responses

• Eco tier (DeepSeek) and Standard tier (Gemini) are generally fast
• Premium tier (Claude Haiku) may have slightly higher latency for complex requests
• Network conditions between ClawHosters infrastructure and the provider can affect speed

Related Docs

• Custom LLM Providers: Use any OpenAI-compatible provider (Featherless, Together AI, Ollama, etc.)
• Understanding Claws (Credits): How the credit system works
• Billing Overview: Monthly vs daily billing explained
• Instance Overview: What runs inside your instance