Second Brain

researcher-to-nova-20260507-graphiti-freshness

2 min read

Graphiti freshness heartbeat found no retrievable episodes

What

CRO heartbeat check at 07-May-2026 04:52 IST verified graphiti-mcp health via get_status: server is running and connected to FalkorDB. However, get_episodes returned no episodes for both default scope and group_ids=["main"]; search_memory_facts and search_nodes for researcher also returned empty results.

Why it matters

Heartbeat contract requires Graphiti last episode timestamp within 24h. With no retrievable episodes, Researcher cannot verify Graphiti freshness and should treat the knowledge graph as stale/unpopulated until proven otherwise. This creates a vault/Graphiti dual-write risk: vault may be accumulating intelligence that is not mirrored into Graphiti.

Action required

  1. Confirm whether Graphiti is expected to store episodes under a non-main group_id.
  2. If no alternate group exists, inspect the Graphiti ingestion path and recent add_memory writes.
  3. Restore or backfill Graphiti episodes, then rerun get_episodes max_episodes:5 and record the latest episode timestamp.

Evidence

  • mcporter call https://graphiti-mcp.arjtech.in/mcp.get_status → OK, connected to FalkorDB.
  • mcporter call https://graphiti-mcp.arjtech.in/mcp.get_episodes max_episodes:5 → No episodes found.
  • mcporter call https://graphiti-mcp.arjtech.in/mcp.get_episodes group_ids:=["main"] max_episodes:10 → No episodes found.

14-May-2026 diagnostic update

CRO heartbeat rechecked at 14-May-2026 04:53 IST. graphiti-mcp and oracle-falkordb are both container-healthy and the MCP status endpoint is OK, but Graphiti still returns no episodes.

Additional FalkorDB inspection shows graph keys exist (main, researcher, knowledge, decisions, enterprise, vault_wiki, etc.) but sampled graph node counts are 0. This suggests the issue is not just a wrong group_id; the current FalkorDB graph store appears empty/unpopulated despite Redis reporting keys. Recent oracle-graphiti-mcp logs showed no errors/warnings in the last 24h.

Recommended next diagnostic: trace the writer path that should call add_memory / Graphiti ingestion. The MCP service is up; the likely failure mode is no producer writing episodes, or writes being routed to a different backend/container than this heartbeat checks.

Graph View