WOLFRAM

Wolfram Language & System Documentation Center

LLMGraph["name1"fun1,…]

creates an LLMGraph where node "namei" evaluates funi on parent nodes' outputs.

LLMGraph[…][input]

gives the result of running the graph on input.

LLMGraph[…][input,prop]

gives the property prop after running the graph on input.

Details
Details and Options Details and Options
Examples  
Basic Examples  
See Also
History
Cite this Page

LLMGraph

LLMGraph["name1"fun1,…]

creates an LLMGraph where node "namei" evaluates funi on parent nodes' outputs.

LLMGraph[…][input]

gives the result of running the graph on input.

LLMGraph[…][input,prop]

gives the property prop after running the graph on input.

Details

  • An LLMGraph can be used to efficiently schedule and combine multiple LLM generation steps for an optimal evaluation managing LLM submissions' concurrency.
  • LLMGraph requires external service authentication, billing and internet connectivity.
  • Possible values for node evaluation function funi include:
  • LLMFunction[…]an LLMFunction for LLM submission
    Function[…]a pure function for Wolfram kernel submission
    <|key1val1, …|>an Association with detailed node specifications nodespec
  • Possible keys in nodespec for evaluation specifications are:
  • "NodeFunction"arbitrary function to evaluate on the node
    "LLMFunction"LLM evaluation via an LLMFunction
    "ListableLLMFunction"threaded LLM evaluation on list input values
    "Input"explicit list of nodes required as function arguments
    "TestFunction"whether the node should run
    "TestFunctionInput"explicit list of nodes required as test arguments
  • Each node must be defined with only one of "NodeFunction", "LLMFunction" or "ListableLLMFunction".
  • The function specified in "ListableLLMFunction" threads over list input. Non-list input is copied. The output of the node is always a list.
  • Possible "LLMFunction" specifications as prompti include:
  • "text"static text
    {"text1",…}a list of strings
    LLMPrompt["name"]a repository prompt
    StringTemplate[…]templated text
    TemplateObject[…]template for creating a prompt
    LLMFunction[…]an LLMFunction object
  • If a prompti is explicitly dependent on the parameter TemplateSlot["namej"], the results from nodej evaluation will be computed first and then used to build the new prompt.
  • If nodej has no function specification, its result can be provided in input.
  • The computation in a node is started as soon as all its dependencies are available.
  • Cyclic prompt dependencies are not supported.
  • Possible values for input are:
  • valvalue for the single graph input
    <|name1val1, …|>values for each of the nodes or inputs namei
  • Possible values for prop are:
  • Automaticresults for the output nodes
    Allresults for all the nodes
    "LLMGraph"a new LLMGraph annotated with the result
  • LLMGraph supports the following options:
  • AuthenticationAutomaticfallback authentication
    LLMEvaluator$LLMEvaluatorfallback LLM configuration
  • To evaluate a node, its own LLMConfiguration nconf will be used. LLMEvaluatorgconf can be used to specify defaults for Missing[…] or Automatic values.

Examples

Basic Examples  (1)

Define a basic graph:

Wolfram Research (2025), LLMGraph, Wolfram Language function, https://reference.wolfram.com/language/res/LLMGraph.html.

Text

Wolfram Research (2025), LLMGraph, Wolfram Language function, https://reference.wolfram.com/language/res/LLMGraph.html.

CMS

Wolfram Language. 2025. "LLMGraph." Wolfram Language & System Documentation Center. Wolfram Research. https://reference.wolfram.com/language/res/LLMGraph.html.

APA

Wolfram Language. (2025). LLMGraph. Wolfram Language & System Documentation Center. Retrieved from https://reference.wolfram.com/language/res/LLMGraph.html

BibTeX

@misc{reference.wolfram_2025_llmgraph, author="Wolfram Research", title="{LLMGraph}", year="2025", howpublished="\url{https://reference.wolfram.com/language/res/LLMGraph.html}", note=[Accessed: 25-August-2025 ]}

BibLaTeX

@online{reference.wolfram_2025_llmgraph, organization={Wolfram Research}, title={LLMGraph}, year={2025}, url={https://reference.wolfram.com/language/res/LLMGraph.html}, note=[Accessed: 25-August-2025 ]}