Skip to content

Analysis request: codysnider/tagmem (local agent memory with MCP and reproducible benchmarks) #3

@codysnider

Description

@codysnider

Hi, I’d love to request an analysis of my project: https://github.com/codysnider/tagmem
tagmem is a local memory storage and retrieval system for LLM agents.

It is:

  • local-first
  • open source
  • Docker-friendly
  • MCP-compatible
  • focused on tagged, depth-aware memory

A few details that may be relevant for analysis:

  • written in Go
  • supports MCP, CLI, and local Docker-based deployment
  • uses local embeddings
  • current published benchmark default is bge-small-en-v1.5
  • includes reproducible benchmark artifacts and methodology in-repo
  • current published LongMemEval result: 99.0% R@5

What may be interesting to evaluate:

  • architecture and retrieval model
  • memory representation (entries, tags, depth, facts, diary)
  • ingestion / tagging pipeline
  • MCP integration
  • benchmark methodology and reproducibility
  • tradeoffs vs. other agent-memory systems
  • strengths, weaknesses, and where the design is novel or standard

Benchmark docs are here:

If you’re open to it, I’d really appreciate a full analysis. Happy to answer implementation or benchmark questions if useful.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions