Redis semantic cache with redis-py

Build a Redis-backed semantic cache for LLM responses in Python with redis-py and sentence-transformers

This guide shows you how to build a small Redis-backed semantic cache for LLM responses in Python with redis-py and the sentence-transformers library. It includes a local web server built with the Python standard library so you can send paraphrased prompts at a mock LLM, watch the cache decide hit or miss, sweep the cosine-distance threshold, and see the cumulative latency and token savings build up.

Overview

Each cache entry is stored as a single Redis Hash at cache:<id>. The hash holds the original prompt, the LLM's response, the raw float32 bytes of a 384-dimensional embedding of the prompt, and metadata fields — tenant, locale, model version, safety flag — plus a created_ts and a hit_count. A single Redis Search index covers the embedding field and every metadata field, so one FT.SEARCH call with a KNN clause does the vector lookup and the TAG pre-filter in the same round trip — no cross-store joins.

The lookup is thresholded: FT.SEARCH always returns the nearest entry that satisfies the filters, but the application only serves it as a hit when the reported cosine distance is at or below distance_threshold. Anything further away is treated as a miss; the caller runs the LLM and writes the new prompt, response, and embedding back to the same key pattern with a TTL.

That gives you:

• A single round trip for lookup — vector KNN + metadata pre-filter in one FT.SEARCH.
• Tens of milliseconds on a hit vs. a multi-second LLM call on a miss; the embedding step is the bottleneck either way, and that's a model-side cost, not a Redis one.
• Tenant, locale, and model-version isolation enforced inside the query, not in application code — a write under one tenant cannot be served to another.
• Bounded memory: every entry has an EXPIRE TTL, and a database-level eviction policy (LRU / LFU) caps the cache size under pressure.

How it works

A query goes through three stages: embed, lookup, and (on a miss) call the LLM and write back.

Hit path (the goal)

1. The application calls embedder.encode_one(prompt) to turn the incoming text into a 384-dimensional float32 vector.
2. cache.lookup(query_vec, tenant=..., locale=..., model_version=...) runs FT.SEARCH with a TAG pre-filter and a KNN 1 clause. Redis returns the closest cached prompt that satisfies the filters along with its cosine distance.
3. If the distance is at or below the threshold, the cache returns a CacheHit containing the cached response. The helper also pipelines an HINCRBY on hit_count and an EXPIRE refresh, so a frequently used answer keeps its TTL and the demo UI can see which entries are load-bearing.
4. The LLM is not called at all. The application returns the cached response to the user.

Miss path

When the distance is above the threshold — or there is no candidate in scope at all — the helper returns a CacheMiss instead, carrying the distance of the nearest candidate (if any) for logging. The application then:

1. Calls the LLM with the prompt.
2. Calls cache.put(prompt, response, embedding, tenant=..., locale=..., model_version=...). The same embedding the lookup used is reused — no re-encode. The helper writes the Hash with HSET and an EXPIRE TTL in a pipeline.
3. Returns the LLM's response to the user. The next semantically similar prompt under the same metadata scope will be a hit.

The cache helper

The RedisSemanticCache class wraps the Redis Search index and the lookup / write flow (source):

import redis
from cache import RedisSemanticCache, CacheHit, CacheMiss
from embeddings import LocalEmbedder

# Use decode_responses=False because the embedding field is raw bytes;
# the helper decodes text fields explicitly where it needs them.
r = redis.Redis(host="localhost", port=6379, decode_responses=False)
cache = RedisSemanticCache(
    redis_client=r,
    index_name="semcache:idx",
    distance_threshold=0.5,    # cosine distance, lower = stricter
    default_ttl_seconds=3600,  # one hour
)
embedder = LocalEmbedder()  # sentence-transformers/all-MiniLM-L6-v2

# One-time index setup (idempotent).
cache.create_index()

# 1) Embed the prompt.
prompt = "How do I return an item?"
query_vec = embedder.encode_one(prompt)

# 2) Look up under a metadata scope. The TAG filter and the KNN
#    travel together in one FT.SEARCH.
result = cache.lookup(
    query_vec,
    tenant="acme",
    locale="en",
    model_version="gpt-4.5-2026",
)

if isinstance(result, CacheHit):
    response = result.response
    print(f"hit ({result.distance:.3f}): {response}")
else:
    # 3a) Miss — call the LLM. (Use your real client here.)
    response = call_llm(prompt)

    # 3b) Cache the new entry. Reuses the same embedding bytes the
    #     lookup used, so we don't pay the encoder twice.
    cache.put(
        prompt=prompt,
        response=response,
        embedding=query_vec,
        tenant="acme",
        locale="en",
        model_version="gpt-4.5-2026",
    )

Data model

Each cache entry is one Redis Hash. The vector field is raw little-endian float32 bytes — no JSON wrapping — because the Redis Search vector encoding expects exactly that.

cache:7c3f8a1b9e02
  prompt=How do I return an item?
  response=You can return any unworn item within 30 days...
  tenant=acme
  locale=en
  model_version=gpt-4.5-2026
  safety=ok
  created_ts=1715990400.123
  hit_count=4
  embedding=<384 × float32 little-endian bytes>

The Redis Search index schema treats every field as queryable in its natural type:

FT.CREATE semcache:idx
  ON HASH PREFIX 1 cache:
  SCHEMA
    prompt         TEXT
    response       TEXT
    tenant         TAG
    locale         TAG
    model_version  TAG
    safety         TAG
    created_ts     NUMERIC SORTABLE
    hit_count      NUMERIC SORTABLE
    embedding      VECTOR HNSW 6 TYPE FLOAT32 DIM 384 DISTANCE_METRIC COSINE

The prompt and response TEXT fields aren't used by the cache lookup itself — that's vector-only — but they make it possible to grep the cache by content from redis-cli for debugging or admin tooling.

The query

The lookup is a hybrid query: a TAG pre-filter expression in parentheses, then =>[KNN 1 @embedding $vec]. With DIALECT 2, Redis applies the filter first and KNN-ranks only the matching documents.

FT.SEARCH semcache:idx
  "(@tenant:{acme} @locale:{en} @model_version:{gpt\-4\.5\-2026} @safety:{ok})
     =>[KNN 1 @embedding $vec AS distance]"
  PARAMS 2 vec <384-float32-bytes>
  SORTBY distance
  RETURN 7 prompt response tenant locale model_version hit_count distance
  DIALECT 2

distance is the cosine distance (0 means identical, 2 means opposite). The result is sorted ascending, so the top row is the closest candidate. The application inspects distance against the threshold and decides hit or miss in user code — Redis returns the row either way, and treating it as a hit or a miss is a policy decision the cache helper owns, not a server-side filter.

The mock LLM

To make the latency and token savings visible without requiring an API key, mock_llm.py provides a deterministic stand-in (source):

from mock_llm import MockLLM

llm = MockLLM(latency_ms=1500.0)  # one and a half seconds per call
response = llm.complete("What is your return policy?")
# response.response       — the templated answer text
# response.latency_ms     — wall-clock time the call took
# response.total_tokens   — estimated prompt + completion tokens

The mock sleeps for the configured latency, then keyword-matches against a small FAQ table to produce an answer. The deliberate slowness is what makes a hit visibly cheaper than a miss in the demo. In production code, you would replace MockLLM with your real client of choice — OpenAI, Anthropic, Bedrock, vLLM, Ollama, anything — without changing the cache helper.

Pre-seeding the cache

In a real deployment the cache fills up organically: a first-time question is a miss, the LLM answers, and the response is written back. For the demo, seed_cache.py pre-loads a small set of canonical FAQ prompts so the very first query lands on a hit (source):

from seed_cache import seed
from cache import RedisSemanticCache
from embeddings import LocalEmbedder

cache = RedisSemanticCache()
embedder = LocalEmbedder()
cache.create_index()
seed(cache, embedder, tenant="acme", locale="en")

The seed list stores the canonical phrasing of each question ("What is your return policy?"). Paraphrases of any of these prompts ("How do I return an item?", "Can I get a refund?") embed close to the canonical entry, so the cache lookup serves the stored response without ever calling the model.

The interactive demo

demo_server.py runs a ThreadingHTTPServer. The HTML page lets you:

• Type a prompt and toggle metadata: tenant, locale, model version. Each combination is a separate cache namespace inside the same index.
• Slide the cosine-distance threshold and see hits flip to misses (and back) on the same prompt, with the actual distance reported on each query.
• Submit with Ask to run the full hit-or-miss path (calls the LLM on a miss, writes the answer back). Submit with Lookup only (no LLM) to sweep the threshold against a fixed prompt without polluting the cache.
• Watch the cumulative panel build up: total queries, cache hits, cache misses, hit ratio, tokens not spent, LLM seconds not waited.
• Inspect every cached entry, including remaining TTL and total hit count, and drop individual entries to simulate eviction.

The server holds one LocalEmbedder, one RedisSemanticCache, and one MockLLM for the lifetime of the process. Endpoints:

Run the demo locally

1. Clone the redis/docs repository and change into the example directory:

   git clone https://github.com/redis/docs.git
   cd docs/content/develop/use-cases/semantic-cache/redis-py
   

2. Install the dependencies:

   pip install redis sentence-transformers numpy
   

3. Make sure a Redis instance with the Redis Search module is running locally on port 6379. Redis Stack or Redis 8 with Search both work.

4. Start the demo server. The first run downloads the all-MiniLM-L6-v2 model (~80 MB) into the local Hugging Face cache:

   python demo_server.py
   

5. Open http://localhost:8085 and try some queries:

   • "What is your return policy?" — exact match against the seed, distance ≈ 0, hit at any threshold.
   • "How fast is delivery?" — paraphrase of the shipping seed; distance around 0.30, hit at the default threshold of 0.5.
   • "How do I return an item?" — slightly looser paraphrase of the returns seed; distance around 0.49, still a hit at the default threshold. Slide the threshold down to 0.4 to see this one flip to a miss.
   • "What payment methods do you accept?" — unrelated to anything in the seed; distance > 0.8, so you'll see a miss, the mock LLM kicks in for ~1.5 s, the new answer is cached, and a follow-up of the same question is now an immediate hit.
   • Switch the Tenant dropdown to globex or initech and re-ask any seeded question — the result flips to a miss because the cache entries live under acme. That's the metadata pre-filter at work inside FT.SEARCH.

   all-MiniLM-L6-v2 puts FAQ-style paraphrases in the 0.3–0.5 cosine-distance range and unrelated queries above 0.8, which is what motivates the 0.5 default. A stricter embedding model (or a domain-tuned one) would let you drop the threshold further; a noisier one would push it up. The right threshold is always a function of the model, the corpus, and how conservative the application needs to be about reuse.

The server is read/write against your local Redis. The default index name is semcache:idx and entry keys live under cache:. Pass --no-reset to keep an existing cache across restarts, --threshold to change the default cosine-distance cutoff, or --llm-latency-ms to make the mock LLM faster or slower for the demo.