API Usage & Credits

Overview

LinkAutocorrect provides the LLM link correction method through an API. If the machine running LinkAutocorrect does not have an internet connection, this link correction method will be unavailable.

The API endpoint should only be called by LinkAutocorrect itself. Use the LMCS API, if you cannot install LinkAutocorrect.

Pricing

Cost calculation

The base cost for an API request is $0.0001 (one credit) for every 500 characters.

The amount of characters is calculated by the accumulating:

  1. Sum of characters from the list of discoverable pages on your website
  2. Amount of characters from the incorrect URL the user has entered
Screenshot from a PC displaying parts of the Paths Overview from the LinkAutocorrect administration panel. The amount of characters used in LLM requests is 1862 with the website having 75 pages in total.

The cost changes depending the on the llm_mode value that you have configured.

llm_mode value Cost multiplier
standard 1x
fast 3x
enhanced 15x
aggressive 15x

A standard API request from a website with 100 pages costs 0.12 cents on average. Given that most 404 errors can be resolved with other link correction methods, this is incredibly cheap.

You will not be charged for failed API requests.

Cost calculation examples

We assume that llm_mode standard is being used.

  • A normal corporate website like netcom-cs.de contains page names totalling about 2,500 characters.
    Realistic usage scenario (1k API requests per month): $0.5
  • Websites with lots of blog articles like nano-control.org contains page names totalling about 15,000 characters.
    Realistic usage scenario (1k API requests per month): $3
  • Large e-commerce sites like oceansapart.com contains page names totalling about 40,000 characters (when only using the page names from one language for the link correction) or 150,000 characters (using page names from all available languages for the link correction).
    Realistic usage scenario (20k API requests per month): $600

The amount of characters the user has entered in the URL is about 65 characters on average, which is negligible.

These sites are the property of their respective owners and are not affiliated with Faray Technologies. These sites have not endorsed the use of LinkAutocorrect.

Response latency

The delay between the API request and the response differs greatly depending on the amount of characters in the API request (see calculation above) and varies between each provider.
The response times should stay below 5 seconds if you keep the llm_provider option set to auto.
We typically see a response time of 2 seconds for llm_mode fast and 4 seconds otherwise.
Response times from our API will never exceed 30 seconds.

Rate-limits

Rate-limits serve multiple purposes and are common across all online services. The most common use for rate-limits is the prevention of brute-force attacks.

We use rate-limits for the following purposes:

  • Protect our service health from being impacted by malicious users
  • Lower the amount of possible damage if a customer loses control over their API key
Rate limit level Limits
Level 1 20 RPM | 150 RPH | 1,000 RPW
Level 2 50 RPM | 500 RPH | 10,000 RPW
Level 3 (default) 100 RPM | 1000 RPH | 10,000 RPD
Level 4 (available upon request) 200 RPM | 5000 RPH | 100,000 RPD
Level 5 500 RPM | 25000 RPH | 500,000 RPD

All faray.tech accounts have rate limit level 3 selected by default. You can manually set stricter rate limits if you prefer to do so. Please Contact us if you need a higher rate limit. After your rate limit level has been set to level 4 or higher you cannot manually lower it.

Requests are counted with a sliding time window. This is different compared to a fixed time window that would reset to zero after the time period is over.

The amount of pages on your website and the selected llm_mode do not affect the rate limit.

Providers

We use external LLM providers to access cost-efficient and the most cutting-edge models.

By default, the llm_provider option is set to auto which ensures that all API requests are answered successfully, even if a provider is experiencing an outage.
You can also manually choose between one of the following providers:

  • OpenAI
  • Google
  • Groq (not to be confused with Grok)

Privacy

We retain a copy of all API requests for 30 days. This serves the following purposes:

  • Contractual accountability - We have to be able to prove that API requsts are answered successfully
  • Response to possible inquires - We have to able to respond if an LLM provider flags content submitted through the API for safety reasons

Our API is not meant for processing personally identifiable information (PII).
Do not create pages on your website like example.com/client/customer-email@example.com.

We never sell user data or share it with third-party companies.