Large Language Model Constants  
 
Constant Value Description
Ports
LLM_PORT_DEFAULT0Use the provider-specific default port.
LLM_PORT_SECURE443Default port for secure HTTPS connections.
Protocols
LLM_PROTOCOL_DEFAULT0Use the provider-specific default protocol.
LLM_PROTOCOL_HTTP1Use the HTTP protocol.
LLM_PROTOCOL_HTTPS2Use the HTTPS protocol.
LLM_PROTOCOL_UNKNOWN-1Unknown or invalid protocol.
Options
LLM_OPTION_NONE0No options specified.
LLM_OPTION_NOCACHE0x000001Disable provider and model caching for the session.
LLM_OPTION_KEEPALIVE0x000002Enable persistent (keep-alive) HTTP connections.
LLM_OPTION_PROXY0x000008Use a configured proxy server for the connection.
LLM_OPTION_NOUSERAGENT0x000020Do not send a User-Agent header with the request.
LLM_OPTION_SECURE0x001000Force use of a secure (HTTPS) connection.
LLM_OPTION_DEFAULTLLM_OPTION_NONEDefault option set (no flags enabled).
Providers
LLM_PROVIDER_NONE0No provider specified.
LLM_PROVIDER_LOCAL1Local inference provider (e.g., LM Studio or Ollama).
LLM_PROVIDER_OPENAI2OpenAI provider using the ChatGPT models.
LLM_PROVIDER_ANTHROPIC3Anthropic provider using the Claude models.
LLM_PROVIDER_GOOGLE4Google AI provider using the Gemini models.
LLM_PROVIDER_MICROSOFT5Microsoft Foundry provider using OpenAI-compatible models.
LLM_PROVIDER_MISTRAL6Mistral AI provider using the Mistral models.
LLM_PROVIDER_UNKNOWN((DWORD)-1)Unknown or unrecognized service provider.
LLM_PROVIDER_DEFAULTLLM_PROVIDER_LOCALDefault service provider.
Reasoning
LLM_REASONING_NONE0No reasoning, even if it is available.
LLM_REASONING_DEFAULT1Provider or model default.
LLM_REASONING_LOW2Prefer faster responses; reduces usage costs.
LLM_REASONING_MEDIUM3Balanced reasoning.
LLM_REASONING_HIGH4Prefer quality responses; increases usage costs.
LLM_REASONING_MAXIMUM5Maximum reasoning quality.
Temperature
LLM_TEMPERATURE_MINIMUM0.0Deterministic output with minimal randomness.
LLM_TEMPERATURE_CONSERVATIVE0.3Low variability, more predictable responses.
LLM_TEMPERATURE_BALANCED0.7Balanced creativity and coherence.
LLM_TEMPERATURE_DEFAULT1.0Standard default sampling behavior.
LLM_TEMPERATURE_CREATIVE1.2Increased creativity and variation.
LLM_TEMPERATURE_MAXIMUM2.0Maximum randomness with highly varied output.
Top-P Sampling
LLM_TOP_P_DISABLED0.0Nucleus sampling disabled.
LLM_TOP_P_MINIMUM0.1Very restrictive token selection.
LLM_TOP_P_CONSERVATIVE0.8Conservative token selection.
LLM_TOP_P_BALANCED0.9Balanced nucleus sampling.
LLM_TOP_P_CREATIVE0.95More diverse token selection.
LLM_TOP_P_MAXIMUM1.0Full probability distribution with no filtering.
LLM_TOP_P_DEFAULTLLM_TOP_P_DISABLEDDefault disables nucleus sampling.
Top-K Sampling
LLM_TOP_K_DISABLED0Top-K sampling disabled.
LLM_TOP_K_MINIMUM0Equivalent to disabled sampling.
LLM_TOP_K_CONSERVATIVE10Restrict to top 10 tokens.
LLM_TOP_K_BALANCED40Balanced token selection range.
LLM_TOP_K_CREATIVE60More diverse token selection.
LLM_TOP_K_MAXIMUM100Wide token selection range.
LLM_TOP_K_DEFAULTLLM_TOP_K_DISABLEDDefault disables Top-K sampling.
Frequency
LLM_FREQUENCY_DISABLED0.0No frequency penalty applied.
LLM_FREQUENCY_MINIMUM0.0Minimum frequency penalty.
LLM_FREQUENCY_MAXIMUM2.0Maximum frequency penalty.
LLM_FREQUENCY_DEFAULTLLM_FREQUENCY_DISABLEDDefault disables frequency penalty.
Presence
LLM_PRESENCE_DISABLED0.0No presence penalty applied.
LLM_PRESENCE_MINIMUM0.0Minimum presence penalty.
LLM_PRESENCE_MAXIMUM2.0Maximum presence penalty.
LLM_PRESENCE_DEFAULTLLM_PRESENCE_DISABLEDDefault disables presence penalty.
Provider Flags
LLM_PROVIDER_FLAG_NONE0No provider requirements specified.
LLM_PROVIDER_FLAG_REQUIRE_TLS0x0001Requires a secure TLS (HTTPS) connection.
LLM_PROVIDER_FLAG_REQUIRE_AUTH0x0002Requires authentication (e.g., API key or token).
LLM_PROVIDER_FLAG_API_VERSION0x0004Requires an explicit API version parameter.
LLM_PROVIDER_FLAG_LOCAL0x0008Locally hosted provider (LM Studio, Ollama, etc.)
LLM_PROVIDER_FLAG_MODEL_LIST0x0010Supports enumerating available models.
LLM_PROVIDER_FLAG_MODEL_INFO0x0020Supports retrieving detailed model information.
Model Options
LLM_MODEL_OPTION_DEFAULT0x0000default list of chat models for a provider.
LLM_MODEL_OPTION_PREVIEW0x0001Include preview or experimental models.
LLM_MODEL_OPTION_STREAMING0x0002Include models that support streaming responses.
LLM_MODEL_OPTION_STRUCTURED0x0004Include models that support structured outputs.
LLM_MODEL_OPTION_REASONING0x0008Include reasoning-oriented models.
LLM_MODEL_OPTION_VISION0x0010Include models that support image input.
LLM_MODEL_OPTION_AUDIO0x0200Include models that support audio output (text-to-speech).
LLM_MODEL_OPTION_IMAGE0x0400Include models that support image generation or input.
LLM_MODEL_OPTION_VIDEO0x0800Include models that support video generation.
LLM_MODEL_OPTION_TRANSCRIPTION0x1000Include models that support speech-to-text transcription.
LLM_MODEL_OPTION_DEPRECATED0x8000Include deprecated or retired models.
LLM_MODEL_OPTION_ALL_MODELS0xFFFFInclude all available model types.
Model Flags
LLM_MODEL_FLAG_NONE0x0000No model capabilities defined.
LLM_MODEL_FLAG_CHAT0x0001Supports conversational text generation.
LLM_MODEL_FLAG_SYSTEM_ROLE0x0002Accepts a system role message in the input.
LLM_MODEL_FLAG_STREAMING0x0004Supports streaming responses.
LLM_MODEL_FLAG_STRUCTURED0x0008Supports structured output (e.g., JSON or schema).
LLM_MODEL_FLAG_VISION0x0010Supports image inputs (multimodal).
LLM_MODEL_FLAG_TOOLS0x0020Supports tool or function calling.
LLM_MODEL_FLAG_REASONING0x0040Supports reasoning-oriented generation.
LLM_MODEL_FLAG_RESPONSES0x0080Supports the Responses API (unified endpoint).
LLM_MODEL_FLAG_FUNCTIONS0x0100Supports OpenAI-style function calling with schemas.
LLM_MODEL_FLAG_AUDIO0x0200Supports audio output (text-to-speech)
LLM_MODEL_FLAG_IMAGE0x0400Supports image generation.
LLM_MODEL_FLAG_VIDEO0x0800Supports video generation.
LLM_MODEL_FLAG_TRANSCRIPTION0x1000Supports speech-to-text transcription.
LLM_MODEL_FLAG_INSTRUCTIONS0x2000Supports a dedicated instruction field outside message lists.
LLM_MODEL_FLAG_MAX_COMPLETION_TOKENS0x4000Uses max_completion_tokens instead of max_tokens in requests.
LLM_MODEL_FLAG_INFERRED0x80000Capabilities are inferred rather than explicitly defined.
Model Status
LLM_MODEL_STATUS_NONE0x0000No status assigned.
LLM_MODEL_STATUS_AVAILABLE0x0001Model is actively available.
LLM_MODEL_STATUS_PREVIEW0x0002Model is in preview or experimental stage.
LLM_MODEL_STATUS_DEPRECATED0x0004Model is deprecated or legacy.
Limits
LLM_MAX_MESSAGE_LENGTH65536Maximum length of a single input message (64 KB)
LLM_MAX_RESPONSE_LENGTH262144Maximum expected response length (256 KB)
LLM_MAX_SYSTEM_PROMPT_LENGTH32768Maximum length of the system prompt (32 KB)
LLM_MAX_CHAT_HISTORY1000Maximum number of messages stored per session.
LLM_MAX_TOKENS200000Maximum token limit per request.
Roles
LLM_ROLE_UNKNOWN0Unknown or invalid role.
LLM_ROLE_USER1User input message.
LLM_ROLE_ASSISTANT2Assistant response message.
LLM_ROLE_SYSTEM3System-level instruction message.
LLM_ROLE_FUNCTION4Function call message (legacy)
LLM_ROLE_TOOL5Tool-generated message.
LLM_ROLE_TOOLUSER6User input directed to a tool.
LLM_ROLE_DEFAULTLLM_ROLE_USERDefault role is user input.
Responses
LLM_RESPONSE_UNKNOWN0No stop reason specified or unrecognized response state.
LLM_RESPONSE_COMPLETE1Response completed normally.
LLM_RESPONSE_LIMITED2Stopped due to token or context limits.
LLM_RESPONSE_ABORTED3Request was cancelled or interrupted.
LLM_RESPONSE_ERROR4An error occurred during processing.
LLM_RESPONSE_TIMEOUT5Request timed out.
Session
LLM_SESSION_FLAG_NONE0No session flags specified.
LLM_SESSION_FLAG_DEFAULTLLM_SESSION_FLAG_NONEDefault session behavior.
Cache
LLM_INVALID_CACHE_TIME-1Indicates an invalid or unspecified cache duration.