Skip to main content
MemoryOS exposes a quota and dependency envelope on every request so your application keeps running even when memory behavior changes.

​
Modes

ModeMeaningTypical behavior
FULLNormal operationReads and writes behave normally
PASSTHROUGHWrites or reads should bypass memory contextYour AI call should continue without MemoryOS context
DEGRADED_RETRIEVERetrieval is degradedRetrieval may return fewer or zero memories
BLOCKEDTenant is blocked from memory writesadd() may be blocked by governance

​
Response headers

MemoryOS sets these headers on live API responses:
HeaderMeaning
X-MemoryOS-Quota-ModeCurrent mode
X-MemoryOS-Budget-RemainingRemaining tenant budget percentage
X-MemoryOS-Quota-ResetNext reset timestamp
X-MemoryOS-Circuit-StatusOverall dependency health
X-MemoryOS-ProcessingWrite-path processing state (normal or delayed)

​
The correct is_passthrough handling pattern

When you retrieve memories: 
result = client.get(
    query=user_message,
    external_user_id="customer-123",
)

if result.is_passthrough:
    prompt_addition = ""
else:
    prompt_addition = result.system_prompt_addition
Key rule:
  • Always make the LLM call
  • only skip MemoryOS prompt context when passthrough is active

​
Webhook events

Operational webhook events currently include:
  • quota.warning
  • quota.critical
  • quota.exhausted
  • quota.reset
  • mode.changed
  • processing.delayed
  • processing.recovered
See Webhooks for payload and signature verification details.