Glossary
Key terms and concepts used in AI and MyTokenGate documentation.
AI Model Terms
LLM (Large Language Model)
A type of AI model trained on vast amounts of text data to understand and generate human-like text. Examples include GPT-5.6, Claude 5, and Gemini 3.
Token
The basic unit of text processing in LLMs. Approximately:
- 1 token ≈ 4 characters in English
- 1 token ≈ 0.75 words
- Token count varies by language
Context Window
The maximum amount of text (in tokens) that a model can process in a single request. Larger windows allow for longer conversations and documents.
Temperature
A parameter (0-2) controlling output randomness:
- 0: Deterministic, consistent outputs
- 0.7: Balanced creativity
- 1.5+: High randomness, creative outputs
Embeddings
Vector representations of text that capture semantic meaning. Used for similarity search, clustering, and retrieval.
Reasoning Model
A model that reasons internally (“thinks”) before answering, such as the GPT-5 series and the thinking modes of Claude and Gemini. Reasoning is slower and costlier but more accurate on hard problems.
Reasoning Effort
A parameter controlling reasoning depth. OpenAI uses reasoning_effort (minimal/low/medium/high, etc.); higher effort means more thinking, and higher latency and cost.
Thinking Budget
Another way to control the amount of reasoning, expressed as a token cap. Anthropic uses budget_tokens; Gemini uses thinkingBudget (0 to disable, -1 for dynamic).
reasoning_content
The field that carries a model’s thinking on OpenAI-compatible endpoints (GLM, DeepSeek, etc.). It is returned separately from the final content and must be preserved verbatim in multi-turn tool-calling flows.
Multimodal
A model’s ability to handle multiple input/output modalities at once, such as text plus images. Vision-language models (VLMs) can understand image content.
API Terms
Chat Completion
An API endpoint for conversational AI interactions. Sends a list of messages and receives a model-generated response.
Streaming
A method of receiving responses piece by piece in real-time, rather than waiting for the complete response.
Function Calling
A feature allowing models to call external functions/tools during response generation.
Rate Limiting
Restrictions on the number of API requests within a time period to prevent abuse and ensure fair usage.
Prompt Caching
Reusing an identical long prefix across requests (system prompt, long documents, tool definitions) to cut cost and time-to-first-token. Cached tokens are billed at a lower cache-read rate.
Structured Output (JSON Schema)
The ability to make a model conform strictly to a given JSON schema. Declared via the json_schema form of response_format, it is a stronger constraint than json_object.
Tool Choice
The parameter controlling whether and how a model calls tools: auto (decides), none (disabled), required/any (must call), or a specific function.
Parallel Tool Calls
A model’s ability to request multiple tools in one response, on by default. Disable with parallel_tool_calls=false (OpenAI) or disable_parallel_tool_use=true (Anthropic).
TTFT (Time to First Token)
The time from sending a request to receiving the first output token. Streaming and prompt caching both improve it.
MyTokenGate Terms
API Key
A unique identifier used to authenticate API requests. Keep it secure and never share it publicly.
Base URL
The API endpoint address: https://gateway.mytokengate.com/v1
Model Router
MyTokenGate’s intelligent routing system that directs requests to the appropriate AI provider.
Billing Meter
Real-time tracking of token usage for cost calculation and invoicing.
Fallback
When a model is configured with multiple suppliers, an unavailable primary supplier automatically switches to a secondary one to preserve availability.
Protocol Terms
OpenAI Protocol
The API format originally defined by OpenAI, now an industry standard. Used by many AI services for compatibility.
Anthropic Protocol
Claude’s native API format (Messages API). Supports advanced features like tool use and extended thinking.
MCP (Model Context Protocol)
An open standard that lets AI clients connect to external tools and data sources in a uniform way. It is implemented by clients (such as Claude Code) and is independent of the model API itself.