LlmClient Control Constants  
 
Constant Value Description
Providers
llmProviderNone 0 No provider specified
llmProviderLocal 1 Local inference provider (e.g., LM Studio or Ollama)
llmProviderOpenAI 2 OpenAI provider using the ChatGPT models
llmProviderAnthropic 3 Anthropic provider using the Claude models
llmProviderGoogle 4 Google AI provider using the Gemini models
llmProviderMicrosoft 5 Microsoft Foundry provider using OpenAI-compatible models
llmProviderMistral 6 Mistral AI provider using the Mistral models
llmProviderUnknown -1 Unknown or unrecognized service provider
llmProviderDefault llmProviderLocal Default service provider
Options
llmOptionNone 0 No options specified
llmOptionNoCache &H0001 Disable provider and model caching for the session
llmOptionKeepAlive &H0002 Enable persistent (keep-alive) HTTP connections
llmOptionProxy &H0008 Use a configured proxy server for the connection
llmOptionNoUserAgent &H0020 Do not send a User-Agent header with the request
llmOptionSecure &H1000 Force use of a secure (HTTPS) connection
llmOptionDefault llmOptionNone Default option set (no flags enabled)
Provider Flags
llmProviderFlagNone 0 No provider requirements specified
llmProviderFlagRequireTls &H0001 Requires a secure TLS (HTTPS) connection
llmProviderFlagRequireAuth &H0002 Requires authentication (e.g., API key or token)
llmProviderFlagApiVersion &H0004 Requires an explicit API version parameter
llmProviderFlagLocal &H0008 Locally hosted provider (LM Studio, Ollama, etc.)
llmProviderFlagModelList &H0010 Supports enumerating available models
llmProviderFlagModelInfo &H0020 Supports retrieving detailed model information
Model Options
llmModelOptionDefault &H0000 Default model filtering (no additional criteria)
llmModelOptionPreview &H0001 Include preview or experimental models
llmModelOptionStreaming &H0002 Include models that support streaming responses
llmModelOptionStructured &H0004 Include models that support structured outputs
llmModelOptionReasoning &H0008 Include reasoning-oriented models
llmModelOptionVision &H0010 Include models that support image input
llmModelOptionAudio &H0200 Include models that support audio output (text-to-speech)
llmModelOptionImage &H0400 Include models that support image generation or input
llmModelOptionVideo &H0800 Include models that support video generation
llmModelOptionTranscription &H1000 Include models that support speech-to-text transcription
llmModelOptionDeprecated &H8000 Include deprecated or retired models
llmModelOptionAllModels &HFFFF Include all available model types
Model Flags
llmModelFlagNone &H0000 No model capabilities defined
llmModelFlagChat &H0001 Supports conversational text generation
llmModelFlagSystemRole &H0002 Accepts a system role message in the input
llmModelFlagStreaming &H0004 Supports streaming responses
llmModelFlagStructured &H0008 Supports structured output (e.g., JSON or schema)
llmModelFlagVision &H0010 Supports image inputs (multimodal)
llmModelFlagTools &H0020 Supports tool or function calling
llmModelFlagReasoning &H0040 Supports reasoning-oriented generation
llmModelFlagResponses &H0080 Supports the Responses API (unified endpoint)
llmModelFlagFunctions &H0100 Supports OpenAI-style function calling with schemas
llmModelFlagAudio &H0200 Supports audio output (text-to-speech)
llmModelFlagImage &H0400 Supports image generation
llmModelFlagVideo &H0800 Supports video generation
llmModelFlagTranscription &H1000 Supports speech-to-text transcription
llmModelFlagInstructions &H2000 Supports a dedicated instruction field outside message lists
llmModelFlagMaxCompletionTokens &H4000 Uses max_completion_tokens instead of max_tokens in requests
llmModelFlagIgnoreTuning &H8000 Model should not use explicit sampling and penalty parameters.
llmModelFlagInferred &H80000 Capabilities are inferred rather than explicitly defined
Responses
llmResponseUnknown 0 No stop reason specified or unrecognized response state
llmResponseComplete 1 Response completed normally
llmResponseLimited 2 Stopped due to token or context limits
llmResponseAborted 3 Request was cancelled or interrupted
llmResponseError 4 An error occurred during processing
llmResponseTimeout 5 Request timed out
Reasoning
llmReasoningDefault 0 Provider or model default
llmReasoningNone 1 Disable reasoning if available
llmReasoningLow 2 Prefer faster responses; reduces usage costs
llmReasoningMedium 3 Balanced reasoning
llmReasoningHigh 4 Prefer quality responses; increases usage costs
llmReasoningMaximum 5 Maximum reasoning quality
Temperature
llmTemperatureMinimum 0.0 Deterministic output (minimal randomness)
llmTemperatureConservative 0.3 Low variability, more predictable responses
llmTemperatureBalanced 0.7 Balanced creativity and coherence
llmTemperatureDefault 1.0 Standard default sampling behavior
llmTemperatureCreative 1.2 Increased creativity and variation
llmTemperatureMaximum 2.0 Maximum randomness (highly varied output)
Top-P Sampling
llmTopPDisabled 0.0 Nucleus sampling disabled
llmTopPMinimum 0.1 Very restrictive token selection
llmTopPConservative 0.8 Conservative token selection
llmTopPBalanced 0.9 Balanced nucleus sampling
llmTopPCreative 0.95 More diverse token selection
llmTopPMaximum 1.0 Full probability distribution (no filtering)
llmTopPDefault llmTopPDisabled Default disables nucleus sampling
Top-K Sampling
llmTopKDisabled 0 Top-K sampling disabled
llmTopKMinimum 0 Equivalent to disabled sampling
llmTopKConservative 10 Restrict to top 10 tokens
llmTopKBalanced 40 Balanced token selection range
llmTopKCreative 60 More diverse token selection
llmTopKMaximum 100 Wide token selection range
llmTopKDefault llmTopKDisabled Default disables Top-K sampling
Frequency
llmFrequencyDisabled 0.0 No frequency penalty applied
llmFrequencyMinimum 0.0 Minimum frequency penalty
llmFrequencyMaximum 2.0 Maximum frequency penalty
llmFrequencyDefault llmFrequencyDisabled Default disables frequency penalty
Presence
llmPresenceDisabled 0.0 No presence penalty applied
llmPresenceMinimum 0.0 Minimum presence penalty
llmPresenceMaximum 2.0 Maximum presence penalty
llmPresenceDefault llmPresenceDisabled Default disables presence penalty
Limits
LlmMaxMessageLength 65536 Maximum length of a single input message (64 KB)
LlmMaxResponseLength 262144 Maximum expected response length (256 KB)
LlmMaxSystemPromptLength 32768 Maximum length of the system prompt (32 KB)
LlmMaxChatHistory 1000 Maximum number of messages stored per session
LlmMaxTokens 200000 Maximum token limit per request