Built-in Extractors
Letta Evals provides a set of built-in extractors that cover the most common extraction needs.
Common Extractors
Section titled “Common Extractors”last_assistant
Section titled “last_assistant”Extracts the last assistant message content.
extractor: last_assistant # Most common - gets final responsefirst_assistant
Section titled “first_assistant”Extracts the first assistant message content.
extractor: first_assistantall_assistant
Section titled “all_assistant”Concatenates all assistant messages with a separator.
extractor: all_assistantextractor_config: separator: "\n\n" # Join messages with double newlinepattern
Section titled “pattern”Extracts content matching a regex pattern.
extractor: patternextractor_config: pattern: 'Result: (\d+)' # Regex pattern to match group: 1 # Extract capture group 1tool_arguments
Section titled “tool_arguments”Extracts arguments from a specific tool call.
extractor: tool_argumentsextractor_config: tool_name: search # Which tool to extract fromtool_output
Section titled “tool_output”Extracts the return value from a specific tool call.
extractor: tool_outputextractor_config: tool_name: searchmemory_block
Section titled “memory_block”Extracts content from a specific memory block.
extractor: memory_blockextractor_config: block_label: human # Which memory block to extractafter_marker
Section titled “after_marker”Extracts content after a specific marker string.
extractor: after_markerextractor_config: marker: "ANSWER:" include_marker: falseNext Steps
Section titled “Next Steps”- Custom Extractors - Write your own extractors
- Extractors Concept - Understanding extractors