TensorAPI.comToken semantics guide
Distinguish AI usage tokens from ledger assets
The phrase AI tokens can refer to language model processing units or to blockchain assets associated with an AI project. These meanings require different identifiers, accounting rules and evidence. A token count reported by an inference API does not describe an asset balance, and a blockchain token’s name does not establish a model capability. Use this guide to make those boundaries visible in your application schema, usage reporting and documentation before building a workflow that touches both domains.
Name which system a field describes
Start with a small terminology map that the whole application follows. Avoid an unqualified tokens field when callers cannot tell whether it means text processing units, integer identifiers or a ledger quantity.
| Concept | Identify it with |
|---|---|
| Language model usage | Model, request attempt and reported count category |
| Tokenized text | Tokenizer identity and ordered vocabulary identifiers |
| Blockchain asset | Chain and contract or mint identifier |
Keep the two accounting systems in separate objects even when one application displays both. An inference result may analyze an asset description, but that does not make its output-token count an asset quantity. Likewise, a tensor of numerical features represents data for computation; it does not become a transferable ledger asset merely because the input concerned tokens.
Read usage according to the provider contract
A model API may report input and output usage, with additional categories depending on the interface. Preserve those fields and their definitions instead of collapsing every number into one unlabeled total. Record the model selection, request attempt, planned allowance and observed stop condition alongside the result. Treat estimates and reported usage as different measurements.
Use counts to investigate context growth, repeated attempts and output that consistently reaches its limit. Do not assume a count alone establishes cost or memory consumption: those depend on the relevant service terms and execution architecture. Keep current commercial calculations outside fixed conceptual examples. Evaluate whether smaller prompts preserve task quality, and keep the complete request representation available through an appropriate execution record when diagnosing a discrepancy.
Keep asset claims connected to evidence
A token interface describes operations and recorded quantities under a particular contract. For example, ERC-20 defines a common Ethereum token interface; it does not establish what an AI-themed project actually delivers. Evaluate a claimed service capability using its technical documentation and observable interface, and record that evidence separately from ledger balances.
When AI classifies an asset description, retain the source, observation time and supplied passages. Use explicit categories such as document type or claimed technical function, with an unknown state when evidence is missing. Preserve precision and unit metadata for numerical ledger values, and avoid joining records by display symbol alone. This gives downstream users a traceable description of the data while keeping language model usage, project claims and asset records conceptually distinct.
Use the official reference for the documented interface; the workflow recommendations above are Tensor API Lab guidance.