An Agent Isn't Recalling Memory Correctly

Why the agent kept responding normally

A memory recall problem never fails a run. The agent simply answers as though it remembers nothing for that workspace, rather than erroring out. That's deliberate: a memory outage shouldn't take an agent offline entirely, but it does mean a conversation can look ordinary while quietly missing context it should have had.

"No recall" vs. "Reduced recall"

Both appear on the platform's own Settings → System dashboard, under Memory Recall, naming exactly which workspaces are affected:

• No recall: memory lookups for that workspace are failing outright.
• Reduced recall: lookups are working but degraded, most often when semantic (vector) search is unavailable and the platform has fallen back to text-only matching.

What to check

1. Embedding provider: if vector search dropped to text-only, the embedding provider connected under Settings → Embedding is the most likely cause. Check its connection status and re-verify the key; see Embedding Provider for what each provider needs.
2. Ollama, if that's your embedding provider: a local Ollama instance that's stopped or unreachable produces exactly this symptom. Confirm it's running and reachable from this instance.
3. Database health: check Settings → System → Storage Health. Recall depends on the same database everything else does; a storage or connectivity problem there shows up here too.

Related

• Memory: the three memory types and how recall normally works
• Embedding Provider: connecting or fixing the provider that powers vector search
• Team and Enterprise Settings: where organization-wide health signals like this one live