Skip to content

LLM Bridge Activity

The LLM Bridge activity connects a process to a Large Language Model (LLM) provider, sending a prompt together with optional role, context, and domain-knowledge instructions, and returning the model's response for use in subsequent workflow steps. This activity is the foundation for adding AI Agent style capabilities to a Robusta RPA process.

Field Description Requirement
LLM provider The AI provider to send the request to (e.g., OpenAI, Anthropic, Azure OpenAI). Required
Api key The API key used to authenticate with the selected LLM provider. Required
LLM model The specific model to use for the request (e.g., gpt-4o, claude-sonnet). Required
Agent role Instructions describing the persona or role the model should adopt when responding. Optional
System context Background information or system-level instructions that frame the model's behavior for the request. Optional
Domain knowledge Additional reference information the model should take into account when generating a response. Optional
Prompt The main instruction or question sent to the model. Required
Attachment The file path of a document or image to send alongside the prompt. Optional
Base64 A Base64-encoded file to send alongside the prompt, as an alternative to Attachment. Optional
Temperature Controls the randomness of the model's response; lower values produce more deterministic output. Optional
Save response as json? A boolean option to parse and store the model's response as a JSON object rather than plain text. Optional
Response variable name The name of the variable where the model's response will be stored. Required

Action Types & Examples

Prompt Execution

  • Format: string
  • Example Result: "Summarize the attached invoice and extract the total amount."

JSON Response

  • Format: boolean
  • Example Result: true

Implementation Examples

Field Setup - LLM provider: OpenAI - LLM model: gpt-4o - Api key: ${openAiApiKey} - Agent role: You are an invoice processing assistant. - Prompt: Summarize the attached invoice and extract the total amount. - Attachment: C:\Robusta\invoice.pdf - Response variable name: llmResult

Execution Parameters - Temperature: 0.2 - Save response as json?: true

Technical Notes

Agent role, System context, and Domain knowledge are all optional but let a single activity be reused for different assistant behaviors without changing the Prompt itself. Provide either Attachment or Base64, not both, when the request needs to include a file. Enabling Save response as json? requires the model to return well-formed JSON; the activity does not reformat a non-JSON response.