mempalace.auto_query.depth_cache

Source: mempalace/auto_query/depth_cache.py

TTL cache for the periodic depth-refresh injection.

The depth signal fires a deterministic query ("session context <wing>") whose daemon round-trip costs most of a second — well over the hook latency budget — while its results barely change within a session. Repeat fires are served from this small on-disk cache; only the first fire per TTL window pays the daemon call.

Failure posture: every path fails open (a broken/corrupt/unwritable cache just means the daemon is queried as before).

Functions

default_cache_path

def default_cache_path() -> str

Cache location: env AUTO_QUERY_DEPTH_CACHE_PATH > default.

cache_key

def cache_key(query: str, wing: str, limit: int) -> str

Stable key over the fields that define a depth query.

load_cached_injection

def load_cached_injection(key: str, ttl_seconds: int, cache_path: Optional[str] = None) -> Optional[str]

Return the cached injection for key if younger than ttl_seconds.

store_injection

def store_injection(key: str, injection: str, cache_path: Optional[str] = None) -> None

Store injection under key, pruning stale siblings. Fails open.