Blog draft: From Storage to Intelligence — The v10.55→v10.57 Arc #929
Replies: 3 comments
-
|
Thanks @filhocf — this is a solid draft and captures the arc well. Three corrections before I'd cross-post it anywhere: 1. Issue numbers are wrong 2. Attribution for entity extraction and insight cards 3. Locale plugins — let's revisit Fix those three and the post is good to go. Happy to cross-post to dev.to once they're addressed. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the review! Corrections applied: 1. Issue numbers fixed:
2. Attribution corrected:
3. Locale plugins:
Updated "What's Next" section:
Updated footer: Contributors: @filhocf (entity extraction #868, insight cards #869, tag_match #904, session chunking #912), @doobidoo (maintainer, squash-merge + conflict resolution, ongoing maintenance v10.55.1→v10.57.3) Ready for cross-post once you confirm these are good. |
Beta Was this translation helpful? Give feedback.
-
|
All three corrections land well — attribution and issue numbers are now accurate. One small update before you post: the "What's Next" section is already behind. Since this morning:
Once those two lines are updated the post is ready. Go ahead and cross-post to dev.to. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hey @doobidoo — you mentioned in #870 that you'd review a blog post if I drafted one around the v10.55→v10.57 arc. Here it is.
Feel free to edit, cut, or redirect. Once you're happy with it, I can cross-post wherever makes sense (dev.to, GitHub blog, etc.).
From Storage to Intelligence: The v10.55→v10.57 Arc in mcp-memory-service
The Problem
Most agent memory implementations stop at "store and retrieve." You save a fact, you search for it later. But as your memory corpus grows past a few hundred entries, raw retrieval isn't enough — you need the system to understand what it knows.
Over the past three releases (v10.55.0 → v10.57.3), mcp-memory-service crossed that threshold: from a storage layer to an intelligence tier.
What Shipped
Entity Extraction (v10.55.0, PR #868)
The memory service now automatically identifies entities (
@mentions,#tags, project names) in stored memories and links them in a knowledge graph. This enables queries like "show me everything connected to @doobidoo" or "what decisions reference #authentication" — without manual tagging.The design is deliberately conservative: regex-based NER (no LLM required), opt-in via
MCP_ENTITY_EXTRACTION_ENABLED, integrated into the existingmaintaincycle as Step 5.Insight Cards (v10.55.0, PR #869)
After entity extraction runs, a second pass detects patterns, trends, and gaps across your memory corpus:
These surface as
memory_type='insight'entries — searchable like any other memory, but generated autonomously.tag_match AND/OR Filtering (v10.55.0, PR #904)
A small but high-impact change:
memory_searchandmemory_listnow accepttag_match="all"for AND filtering. Previously, searching fortags=["project-x", "decision"]returned anything tagged with either — now you can require both. Essential for multi-project workflows.Session Chunking (v10.56.0, PR #912)
memory_store_sessionnow automatically chunks long conversations (>4000 chars) into semantically coherent pieces. Each chunk is independently searchable while maintaining session-level coherence viasession:<id>tags.Real-World Numbers
From a production deployment (~2,700 memories, sqlite-vec backend, 3 machines synced):
@mentions/#tags, useful for cross-referencing projects across sessionsWhat's Next
Try It
Or with entity extraction + insight cards enabled:
Contributors: @doobidoo (maintainer, entity extraction, insight cards), @filhocf (tag_match, session chunking, harvest locale plugins)
Beta Was this translation helpful? Give feedback.
All reactions