Degradation Modes - MemoryOS

MemoryOS exposes its current operating mode on every response so your app keeps running regardless of what’s happening with memory.

Modes

Mode	Meaning
`FULL`	Normal — reads and writes work as expected
`PASSTHROUGH`	Memory context should be skipped — continue the LLM call without it
`DEGRADED_RETRIEVE`	Retrieval is reduced — fewer or zero memories may return
`BLOCKED`	Writes are blocked by quota governance

Response headers

Header	Meaning
`X-MemoryOS-Quota-Mode`	Current mode
`X-MemoryOS-Budget-Remaining`	Remaining budget as a percentage
`X-MemoryOS-Quota-Reset`	Next reset timestamp
`X-MemoryOS-Circuit-Status`	Overall dependency health
`X-MemoryOS-Processing`	`normal` or `delayed`

Handling passthrough

from memoryos import MemoryOSError

try:
    result = client.get(
        query=user_message,
        external_user_id="customer-123",
    )
    prompt_addition = result.system_prompt_addition if result.has_context else ""
except MemoryOSError:
    # Record the failure in your server telemetry, then continue without memory.
    prompt_addition = ""

# Always call your model with the user's message.

The rule is simple: always make the model call. Skip MemoryOS context when passthrough is active, retrieval is empty, or the memory request fails. Do not swallow authentication and configuration failures during startup checks; the runtime fallback is for keeping an already-running customer conversation available.

Webhooks

Operational events include quota.warning, quota.critical, quota.exhausted, quota.reset, mode.changed, processing.delayed, and processing.recovered. See Webhooks for payload format and signature verification.