Alternative forms of memory (e.g. Knowledge Graphs)

[Crunchyman-ralph]

Discussed in https://github.com/eyaltoledano/claude-task-master/discussions/487

^{Originally posted by BradKML May 13, 2025} There are some memory designs that are different from just chat history (https://github.com/eyaltoledano/claude-task-master/discussions/486) and conventional memory banks (https://github.com/eyaltoledano/claude-task-master/discussions/284) that might be worth looking into for documentation and problem-solving.

Regarding knowledge graphs and PKM integrations, here are some examples:

• https://github.com/basicmachines-co/basic-memory
• https://github.com/shaneholloman/mcp-knowledge-graph
• https://github.com/okooo5km/memory-mcp-server
• https://github.com/gannonh/memento-mcp
• https://github.com/T1nker-1220/memories-with-lessons-mcp-server

There are some examples where knowledge graphs can probably do better than just vector DB or fuzzy text search:

1. Capturing lists of errors and failed solutions, to reflect on common issues and discover prevention or shared causes (DAGs mixing divergent thinking and convergent thinking)
2. Clustering related features together, for those with tightly coupled dependencies, and slowly adopting proper design patterns to decouple them effectively (node distances and community detection)
3. Creating new ideas by linking user requirements together in novel ways like "deep research" (web crawling) and other forms of lateral thinking

Some questions to consider:

1. How can MCPs interact with one another (task master vs knowledge graphs vs deep research) without clashing in the process?
2. Out of those links, which ones should be selected to be the default to work with Task Master? Basic Memory or MCP-KG?