procedural-memory

What problem does it solve?

Procedural-memory automatically turns raw episodic traces into reusable, verifiable procedures so an agent can recall and execute multi-step workflows without manual curation.

Core Features & Use Cases

• Procedure extraction from episodic traces: mines recurring ordered action patterns from continuity logs, transcripts, or execution traces.
• LLM-based knowledge verification: filters frequent patterns into high-quality procedural candidates using an explicit scoring rubric.
• Procedural DAG construction with dual-indexing: builds START→steps→GOAL graphs and generates both goal-level and step-level embedding indexes for retrieval.
• Incremental EMA updates: updates index vectors and transition statistics as new experiences arrive while reducing catastrophic forgetting.
• Hybrid retrieval via neural + symbolic queries: supports semantic discovery (by goal/step similarity) and structured DAG queries (prerequisites, next steps, alternative paths).

Quick Start

Use procedural-memory to extract and index recurring workflows from your session execution traces, then query the resulting procedural DAG for the steps required to reach a goal.