ShinrAIHosted on STACKIT
Understand your usage

What is a record?

A record measures how much text ShinrAI processes. It is not a user, a file, an AI token or a monthly seat.

Start with 1,000 characters

With Standard processing, a text up to 1,000 characters uses one record. A 2,500-character text uses three. Each text is rounded separately.

We count Unicode code points, including spaces and punctuation. A displayed character can contain several code points. UTF-8 bytes and AI tokens are different units.

Small examples, clear totals

Records used by the same text at each processing tier
Submitted textStandardBatchReal-Time
1,000 characters10.51.6
1,001 characters213.2
2,500 characters31.54.8
Two separate 500-character texts213.2

Sending two texts in one request does not combine their rounding. Supported empty documents still have a minimum charge of one block; invalid requests are rejected.

A company chatbot example

A 1,000-character question plus four 1,500-character retrieved passages makes one 7,000-character submission: seven Standard records. At 8,000 questions per month, that is 56,000 records.

Indexing 1,000 new files of 3,000 characters each with Batch adds 1,500 records. The combined monthly estimate is 57,500. Initial indexing of an existing library is a separate, one-time workload.

Conversation history, retrieved emails and repeated agent steps count again when submitted again. Protecting a document during indexing does not make later API requests free. Restoring values locally does not itself use another ShinrAI record.

Allowance, top-ups and optional help

Your subscription month starts on your billing date. Unused monthly records do not carry over. Purchased packs do not expire.

Monthly allowance is used before managed record packs. Offline packs belong to one offline installation and are separate from managed API usage. Partner onboarding and training are optional one-time services, not record allowances.

Check usage.weighted_records in the API response and the remaining balance in Dashboard. Measure a representative workload before committing to a larger rollout.