Optional systemSystem-level prompts with optional priority ordering. Higher priority usually means placed earlier or given more weight.
Optional priority?: numberOptional source?: stringOptional conversationConversation history messages, typically an array of Message objects.
Optional currentHow many messages at the end of conversationHistory belong to the turn
being answered: the current user message when it travels as history
(structured input), then the assistant tool calls and tool results of
the turn's earlier calls. Token budgeting summarizes or truncates only
the messages before them, so the model always receives the request it
answers and its own tool rounds. A tool round the first of them belongs
to is kept whole with them: a call before them whose result they hold is
not dropped. Defaults to 0.
Optional userCurrent user input text.
Optional visionVisual inputs (images) if the target model supports vision.
Optional audioAudio input data references if the model or a pre-processing step handles audio.
Optional toolsAvailable tools and their schemas, for models that support function/tool calling.
Optional toolPre-formatted tool/function schemas that should be forwarded to the model as-is. Useful when upstream logic has already normalized the schema definitions.
Optional retrievedRetrieved context from a RAG system, to be incorporated into the prompt.
Optional assembledAssembled cognitive memory context (personality-affected, token-budgeted). Merged with retrievedContext by the prompt template.
Optional taskTask-specific data or parameters that need to be included in the prompt.
Optional customAdditional custom components that templates might use.
Core components that form the foundation of any prompt construction. These are gathered from various sources (GMI state, user input, RAG) and then augmented with dynamically selected contextual elements.
Interface
PromptComponents