Configuration
| Setting | Description |
|---|---|
| Integration | The LLM provider (Gemini, OpenAI, Anthropic) |
| Prompt Template | The instruction sent to the LLM. Use Mentions to inject record values dynamically |
| Data Fields | Additional structured data sent alongside the prompt |
| Result Field Name | The name of the output field that stores the LLM’s response |
| Temperature | Controls response creativity (0 = deterministic, 1 = creative) |
| Model | The specific model to use (e.g., Gemini 2.5 Flash Lite) |
| Limit | Maximum number of requests to process |
How It Works
Write a prompt template
Write the instruction for the LLM. Use Mentions to inject record values (e.g.,
{{Account Name}}, {{Industry}}).Output
A new field (named by Result Field Name) is added to each record containing the LLM’s response. This field becomes available as a Mention in all downstream nodes.Example
Enrich Salesforce Accounts with employee count estimates:- Set the prompt to instruct the LLM to return an employee count
- Use Mentions to pass
Account Name,Industry, andEmployeesas context - Set the result field to
num_employees - The enriched field becomes available as a Mention in downstream nodes
Best Practices
- Be specific in your prompt template — vague prompts produce inconsistent results
- Use low temperature (0–0.2) for factual extraction, higher for creative tasks
- Set a Limit when testing to avoid processing the entire dataset
- Include relevant context fields in Data Fields to improve LLM accuracy
Related Nodes
- Data Normalization — uses an LLM to clean existing fields rather than generating new ones
- Web Search — enriches records with live web data instead of LLM-generated content