mm-memory-writer

The memory writer captures and indexes episodic, semantic, and procedural memories from completed pipeline runs into an MCP memory server. It handles memory creation, deduplication via search, and updates to superseded memories, tagging them with ticker, sector, and run details.

In automated research pipelines, episodic memories capture run-level events and decisions, semantic memories persist durable beliefs and patterns across runs, and procedural memories record recurring workflows and validation checks to guide future analyses.

Memory creation handles deduplication by performing a search against existing stored memories before indexing new ones. If a match is found, the memory writer updates the superseded memories rather than creating duplicates, ensuring the semantic store remains accurate.

Yes, the memory writer operates without external dependencies to store structured memories. It directly processes completed pipeline outputs and handles memory creation, deduplication via search, and updates to superseded memories with meta-tags like ticker and sector.

Yes, you can persist durable beliefs and patterns across pipeline runs using semantic memory. The memory writer extracts these durable patterns from completed runs and stores them in the MCP memory server for long-term recall and continuity.

When indexing memories from pipeline analysis, the memory writer applies meta-tags including ticker, sector, and run details. These structured tags enable analysts to later query and filter episodic, semantic, and procedural memories effectively.