RequestConfig

Per-request override dataclass for TTL, semantic threshold, cache bypass, and tags. Use when different requests in the same application need different cache behaviour.

Fields

Field	Type	Default	Description
ttl	int | None	None	Override TTL in seconds for this request. None = use TTLPolicy default
semantic_threshold	float | None	None	Override cosine similarity threshold. None = use global threshold
skip_cache	bool	False	Bypass cache entirely — no read, no write
tags	list[str]	[]	Additional invalidation tags for this cache entry
namespace_prefix	str | None	None	Extra key prefix for per-request tenant isolation

---

Usage

from chengeta_ai.core.request_config import RequestConfig

Short TTL for volatile data

cfg = RequestConfig(ttl=30)  # cache for only 30 seconds

# Apply manually in your cache logic:
if not cfg.skip_cache:
    cached = cache.get(messages, model_id="gpt-4o")
if cached is None:
    result = llm_fn(messages)
    cache.set(messages, result, model_id="gpt-4o", ttl=cfg.ttl)

Skip cache for debugging or forced refresh

cfg = RequestConfig(skip_cache=True)

if cfg.skip_cache:
    result = llm_fn(messages)  # always live
else:
    result = cache.get_or_generate(messages, llm_fn)

Lower semantic threshold for broader matching

cfg = RequestConfig(semantic_threshold=0.80)

# Use with AdaptiveSemanticCache or SemanticCache manually:
if not cfg.skip_cache:
    result = sem_cache.get(query)  # uses cache's default threshold
    # For per-request threshold, pass to vector backend directly

Add extra invalidation tags

cfg = RequestConfig(tags=["session:abc123", "user:42"])
cache.set(messages, result, model_id="gpt-4o", tags=cfg.tags)

# Later invalidate by session:
manager.invalidate("session:abc123")

---

Design Note

RequestConfig is a plain dataclass — it carries intent, not logic. Apply it in your own cache orchestration layer or pass it through middleware. Future versions of CacheManager.get() and ResponseCache.get_or_generate() will accept RequestConfig directly.