Add LangCache wrapper (LLM cache extension) with configurable distance scale

[abrookins]

Summary

Add a LangCache wrapper to RedisVL LLM cache extensions so users can leverage LangCache for semantic caching via the familiar BaseLLMCache interface. Also make the wrapper’s distance threshold semantics explicit and configurable so users can swap it out for an existing SemanticCache instance easily.

Motivation

• Use LangCache while keeping RedisVL’s developer experience
• Avoid surprises by making threshold scale explicit

What’s included

• New LangCacheWrapper implementing BaseLLMCache (sync + async) that talks to the LangCache service:
  • Exact + semantic search toggles, TTL, and standard store/check APIs
  • Consistent result shape (including vector_distance) compatible with existing callers
  • Lazily imports LangCache and gives a friendly error if it's not installed
• Configurable distance threshold scale via distance_scale:
  • "normalized": interpret distance_threshold as 0–1 semantic distance; converts to/from similarity (1 - d)
  • "redis": interpret distance_threshold as native COSINE [0–2]; converts using norm_cosine_distance / denorm_cosine_distance

Usage

from redisvl.extensions.cache.llm.langcache import LangCacheWrapper

cache = LangCacheWrapper(
    name="langcache",
    cache_id="<id>",
    api_key="<key>",
    distance_scale="normalized"  # or "redis" for COSINE [0–2]
)

cache.store(prompt="Hello?", response="Hi\!")

hits = cache.check(
    prompt="Hello\!",
    distance_threshold=0.15,  # interpreted per distance_scale
)

Docs

• Follow-up docs PR will add LangCache wrapper docs and clarify threshold scales

Related

• Opened SemanticCache threshold scale inconsistencies (async normalize_vector_distance and docstring) #407 to track threshold-scale inconsistencies (docstring vs sync/async) in SemanticCache

[Copilot]

Pull Request Overview

This PR adds a LangCache-based LLM cache wrapper with configurable distance scale and fixes an async connection parameter in AsyncSearchIndex. It also introduces unit/integration tests and updates packaging to include an optional langcache extra.

• Add LangCacheWrapper with exact/semantic search, consistent hit shape, and distance scale support ("normalized" or "redis")
• Fix AsyncSearchIndex.connect to pass url kwarg to the connection factory; add integration test
• Update packaging: add langcache extra; adjust Python requirement

Reviewed Changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/unit/test_langcache_wrapper.py	New unit tests for LangCacheWrapper covering init, store/check paths, threshold conversion, and delete/update behavior.
tests/integration/test_async_search_index.py	Adds regression test ensuring AsyncSearchIndex.connect accepts redis_url and connects successfully despite deprecation.
redisvl/index/index.py	Fixes async connect to call RedisConnectionFactory with url=redis_url (consistent with factory’s API).
redisvl/extensions/cache/llm/langcache.py	Implements LangCacheWrapper with distance scale conversions, sync/async store/check/delete, and consistent CacheHit output.
redisvl/extensions/cache/llm/init.py	Re-exports LangCacheWrapper in the llm package namespace.
pyproject.toml	Adds optional dependency extra for langcache and tightens the minimum Python version to 3.9.2.

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[abrookins]

Addressed review feedback:

• Broadened attributes type in check/acheck to Optional[Dict[str, Any]]
• Clarified docstrings to document both conversion paths (normalized vs redis)
• Reduced duplication by factoring helpers: _similarity_threshold, _build_search_kwargs, _hits_from_response

[rbs333]

Overall looks good to me! Left a couple comments that I think we should sync on

[abrookins]

Fully deprecated url as a parameter to the async connection functions per review.

[rbs333]

Looks good!

[bsbodden]

LGTM!