Fair Use Policy

Our "Unlimited" subscription is intended for personal, development, and testing workloads, including long coding sessions, IDE integrations, and autonomous agents. This Fair Use Policy ensures consistent performance and fair access for all users.

Scope of Intended Use

The standard subscription is designed primarily for:

• Personal projects
• Local tools and scripts
• IDE integrations
• Agent frameworks (e.g., OpenClaw)
• Development and testing workloads
• Experimentation and research

Note: Heavy long-running workloads such as 24/7 agents are supported within operational limits, provided they don't negatively impact other users or resemble a public/shared API.

Use in commercial products, SaaS platforms, public-facing applications, or services provided to third parties is not included by default. We offer business accounts with multi-seat management and custom pricing.

What Is Included

• Unlimited API requests within fair-use limits
• Access to all models per the model list
• OpenAI-compatible API access
• Swiss-hosted infrastructure with full privacy (no prompt logging or training on your data)
• Hard system limits: 3 parallel requests, 240 requests/min, 10M tokens/min

Prefix Cache & Prefill

Every request is checked against a cache of recently processed conversation prefixes. Tokens that match are served from the cache — fast and cheap. Tokens that do not match have to be re-processed from scratch ("prefilled"), which is the most compute-intensive part of serving a request and the main cost driver on shared infrastructure.

The cache only matches byte-identical leading tokens. To get the most out of it:

• Keep your system prompt and conversation history in a stable order across requests
• Put volatile parts (timestamps, ids, status inserts) at the end of the prompt, not at the front
• Keep the number of concurrent sessions modest

With a stable prefix, workloads typically reach 90%+ cache hit rates — and most of the first-token latency disappears. Cache space is shared and finite: many long-lived sessions at the same time evict each other, turning what could have been cheap cache hits into full re-processing. That is why sustained fleets of parallel sessions are the main reason for fair-use enforcement.

Reasonable Use Guidelines

The following usage patterns are considered reasonable:

• Development and debugging workflows
• Personal AI assistants or tools
• Local automation tasks
• Testing prompts, agents, or integrations
• Moderate batch processing for development purposes

Not Included in Standard Plan

The following require prior approval:

• Public-facing apps or websites
• Customer-facing AI features
• Commercial SaaS integrations
• Reselling or proxying API access
• High-volume production workloads outside typical developer patterns

For business accounts, please contact us.

Monitoring & Enforcement

To maintain reliability, we may monitor activity patterns such as:

• Cache hit rates sustained (20+ minutes) below 85%
• Uncached (re-prefilled) token volumes that make up a disproportionate share of the fleet's compute in a given time window
• Sustained parallel sessions well above the per-key concurrency limit
• Excessive sustained throughput beyond typical developer usage
• Large-scale automated request generation
• Activity resembling a public API or shared backend service

If usage exceeds fair-use expectations, we follow a warning-first approach:

1. We contact you first to understand your workload
2. We suggest a suitable plan or arrangement
3. If the pattern continues, we apply temporary limits to stabilize the system
4. Access may be suspended or terminated in severe or repeat cases

We always aim to communicate and resolve issues before taking restrictive action. In acute cases, where the impact on the fleet is immediate, we may act first and contact you right after — with the numbers that led to the decision.

To keep enforcement responsive, we may use automated or semi-automated measures. If your usage trips a fair-use signal, access to the affected model may be temporarily restricted, and we will email you with the relevant numbers and how to regain access. Once you have resolved the underlying issue, restoring access is usually immediate: generate a new API key from your dashboard — it is active right away, no verification step — or write to support@airouter.ch to have access reinstated on your current key.

Questions

If you're unsure whether your use case fits within fair use, contact support@airouter.ch. We're happy to help you find the right setup.