ReMe/doc/operations_documentation.md

Operations Documentation

This document provides an overview of all operations in the ExperienceMaker framework.

Operations Overview

Op Name	Class	Description	Parameters
Build Query	BuildQueryOp	Builds retrieval query from user request. Extracts query from request.query or constructs it from messages using LLM if enabled.	op.build_query_op.params.enable_llm_build = true/false - Enable LLM-based query construction from messages
Recall Experience	RecallExperienceOp	Recalls relevant experiences from vector store based on the built query. Performs semantic search and retrieves top-k candidates.	op.recall_experience_op.params.retrieve_top_k = 15 - Number of experiences to retrieve op.recall_experience_op.params.query_enhancement = false - Enable query enhancement with message context
Rerank Experience	RerankExperienceOp	Reranks and filters recalled experiences using LLM evaluation and score-based filtering to improve relevance.	op.rerank_experience_op.params.enable_llm_rerank = true - Enable LLM-based reranking op.rerank_experience_op.params.enable_score_filter = false - Enable score-based filtering op.rerank_experience_op.params.min_score_threshold = 0.3 - Minimum score threshold for filtering op.rerank_experience_op.params.top_k = 5 - Number of top experiences to return
Rewrite Experience	RewriteExperienceOp	Generates and rewrites context messages from reranked experiences to make them more relevant and actionable for the current task.	op.rewrite_experience_op.params.enable_llm_rewrite = true - Enable LLM-based context rewriting
Merge Experience	MergeExperienceOp	Merges the list of experiences into a single formatted message that can be used as context in downstream operations.	-
Trajectory Preprocess	TrajectoryPreprocessOp	Preprocesses trajectories by validating their structure and classifying them into success/failure categories based on score thresholds. Sets up context for downstream operations.	op.trajectory_preprocess_op.params.success_threshold = 1.0 - Score threshold to classify trajectories as successful
Trajectory Segmentation	TrajectorySegmentationOp	Segments trajectories into meaningful step sequences using LLM-based analysis. Identifies natural breakpoints based on logical completion, context switches, tool boundaries, and reasoning phases.	op.trajectory_segmentation_op.params.segment_target = "all" - Which trajectories to segment ("all", "success", "failure")
Experience Validation	ExperienceValidationOp	Validates the quality and usefulness of extracted experiences using LLM-based assessment. Evaluates actionability, accuracy, relevance, clarity, and uniqueness of experiences.	op.experience_validation_op.params.validation_threshold = 0.3 - Minimum score threshold for experience validation
Experience Storage	ExperienceStorageOp	Stores validated and deduplicated experiences to the vector database. Converts experiences to vector nodes and handles workspace-specific storage operations.	op.experience_storage_op.params.default_workspace_id = "default" - Default workspace ID when not specified in request
Experience Deduplication	ExperienceDeduplicationOp	Removes duplicate experiences by comparing embeddings of experience content. Performs similarity analysis against both existing stored experiences and current batch experiences.	op.experience_deduplication_op.params.similarity_threshold = 0.5 - Cosine similarity threshold for duplicate detection op.experience_deduplication_op.params.max_existing_experiences = 1000 - Maximum number of existing experiences to compare against
Comparative Extraction	ComparativeExtractionOp	Extracts comparative experiences by analyzing differences between high/low scoring trajectories (soft comparison) and success/failure patterns (hard comparison). Uses similarity analysis to find comparable step sequences.	op.comparative_extraction_op.params.enable_soft_comparison = true - Enable highest vs lowest score comparison op.comparative_extraction_op.params.enable_similarity_comparison = false - Enable success vs failure similarity comparison op.comparative_extraction_op.params.max_similarity_sequences = 5 - Maximum sequences to compare for similarity op.comparative_extraction_op.params.similarity_threshold = 0.3 - Similarity threshold for step sequence matching op.comparative_extraction_op.params.max_similarity_pairs = 3 - Maximum similar pairs to extract experiences from
Simple Summary	SimpleSummaryOp	Generates simple text summaries from individual trajectories by analyzing execution process and results. Creates basic experiences from trajectory completion status.	op.simple_summary_op.params.success_score_threshold = 0.9 - Score threshold to classify trajectory as successful
Success Extraction	SuccessExtractionOp	Extracts actionable experiences specifically from successful trajectories. Processes both segmented step sequences and entire trajectories to identify successful patterns and strategies.	-
Failure Extraction	FailureExtractionOp	Extracts learning experiences from failed trajectories to identify common pitfalls and failure patterns. Processes both segmented sequences and complete trajectories for failure analysis.	-
Update Vector Store	UpdateVectorStoreOp	Updates vector store by inserting new vector nodes or deleting existing ones. Handles batch operations for experience storage and management in workspace-specific vector databases.	-
Recall Vector Store	RecallVectorStoreOp	Retrieves relevant experiences from vector store using semantic search. Filters results by score threshold and removes duplicates based on content similarity.	op.recall_vector_store_op.params.threshold_score = <float> - Minimum similarity score threshold for filtering results (optional)
Vector Store Action	VectorStoreActionOp	Performs administrative actions on vector store workspaces including copy, delete, dump, and load operations. Manages workspace lifecycle and data migration between environments.	Action-specific parameters passed through request context (workspace IDs, file paths, etc.)
React V1	ReactV1Op	Implements ReAct (Reasoning and Acting) agent framework for interactive task execution. Manages tool usage, reasoning steps, and multi-step problem solving with configurable tools and step limits.	op.react_v1_op.params.max_steps = 10 - Maximum number of reasoning/action steps op.react_v1_op.params.tool_names = "code_tool,dashscope_search_tool,terminate_tool" - Comma-separated list of available tools