Skip to content

cianfhoghlaim/cianchosaint / .agents/skills/agent-memory-systems

agent-memory-systems

Unverified


name: agent-memory-systems description: Router for the agent memory capability. Use when an agent needs to remember, recall, search, or learn across sessions. The 5 memory backends are Cognee (knowledge graph), Graphiti (temporal knowledge graph), LanceDB (vector search), FalkorDB (vector + graph hybrid), and Memgraph (production graph). The Cianfhoghlaim-recommended default is Cognee for structured knowledge + Graphiti for temporal + LanceDB for vector RAG. Triggers: 'agent memory', 'cognee', 'graphiti', 'falkordb', 'memgraph', 'lancedb memory', 'cognify', 'remember', 'recall', 'long-term memory', 'episodic memory', 'temporal knowledge graph', 'knowledge graph for agents'.

Agent Memory Systems — Router

The Cianfhoghlaim platform has 5 memory backends. This skill is the router — pick the right one for the task.

The 5 backends

If you need to…Use this backendSkill
Build a knowledge graph from documents (chunks → entities → relationships)Cogneecognee
Track how knowledge changes over time (bi-temporal, episodic)Graphitigraphiti or graphiti-core
Store + search embeddings (vector RAG, hybrid search)LanceDBlancedb
Vector + graph hybrid (Redis-compatible graph queries)FalkorDBfalkordb
Production-grade graph (Cypher, MAGE algorithms, Lab UI)Memgraphmemgraph

Decision tree

Need to remember past conversations + decisions?   → Graphiti (temporal)
Need to build a knowledge graph from documents?     → Cognee
Need hybrid vector + graph in one store?            → FalkorDB
Need pure vector RAG with HNSW?                     → LanceDB
Need Cypher + production graph analytics?           → Memgraph

The Cianfhoghlaim memory stack (the recommended pattern)

For Cianfhoghlaim agents, the canonical pattern is:

  [agent runtime]
       │
       ├──► Cognee  (structured knowledge: entities + relationships)
       │
       ├──► Graphiti  (temporal: when did this fact become true?)
       │
       └──► LanceDB  (vector RAG: similar past context)

Why all 3 (not just 1):

  • Cognee turns documents into a knowledge graph. It handles the "this is about X, related to Y" relationships that vector search can't.
  • Graphiti adds the temporal dimension. "The Leaving Cert syllabus changed in 2024" is true after 2024 and false before. A vector index would return it for any query, regardless of when the question is asked.
  • LanceDB handles the raw vector search. "Find me a chunk that's similar to this question" — no need for a graph.

Why not FalkorDB / Memgraph:

  • FalkorDB is great for Redis-compatible graph + vector in one store. Cianfhoghlaim does not use it in production (the 3 above are the canonical choice). It shines when you want one store, not three.
  • Memgraph is the production graph for teams that need Cypher + MAGE algorithms + a Lab UI. Cianfhoghlaim does not use it for agent memory (Cognee + Graphiti + LanceDB cover the use cases).

Cianfhoghlaim conventions (the rules)

  1. All agent memory calls go through the LiteLLM gateway (http://litellm:4000/v1). Never call the underlying LLM provider directly. This is for cost tracking + Langfuse observability.
  2. All Cognee + Graphiti calls are wrapped in the agents/ shim. The actual model-layer lives in agents/meaisinfhoghlaim/memory/.
  3. All FalkorDB + Memgraph calls go through storage/graph/ (the application-layer client).
  4. LanceDB schemas follow the cocoindex/ v1 App pattern. Do NOT create ad-hoc LanceDB tables.

Pair this skill with

  • cianfhoghlaim-storage/SKILL.md — the Cianfhoghlaim storage mental model (which backends store what)
  • cognee/SKILL.md — the Cognee detail
  • graphiti/SKILL.md — the Graphiti detail
  • lancedb/SKILL.md — the LanceDB detail
  • agent-observability/SKILL.md — how to trace memory calls

Cross-references

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 469kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 321kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 321kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 321kUpdated Claude CodeCodexVerifiedInstall