Getting started

Quickstart

From zero to a semantic cache hit in about five minutes. You need Docker — everything else, including the CLI, ships inside the image. Community edition is free and needs no license.

1. Pull and run#

shell

docker pull crowkis/crowkis:latest

docker run -d --name crowkis \
  -p 127.0.0.1:6379:6379 \
  -p 127.0.0.1:6380:6380 \
  -p 127.0.0.1:6381:6381 \
  -v crowkis-data:/data \
  crowkis/crowkis:latest

Endpoint	Port	What it is
RESP3 (Redis protocol)	127.0.0.1:6379	crowkis cli and any Redis client
Dashboard + REST	127.0.0.1:6380	live verdict feed, management API, /health
gRPC (h2c)	127.0.0.1:6381	protobuf surface

No environment variables are required to boot — defaults are sensible and the data volume persists your cache across restarts. Hardening and every knob: the Docker guide.

2. Talk to it#

The image ships the interactive REPL — crowkis cli — which connects like redis-cli does. Standard Redis commands work, and the C* family adds the semantic layer:

docker exec -it crowkis crowkis cli

PING
CSET "Explain vector caches" "Vector caches store embeddings of past queries so similar questions can reuse answers." EX 86400 MODEL gpt-4o TENANT demo
CGET "Explain vector caches" TENANT demo
CGET "what are vector caches?" TENANT demo
CSIM "France capital city" K 5
CVECCOUNT

The fourth line is the point: a paraphrase of the stored question still hits, because Crowkis matches meaning and structure — not bytes. Already a Redis shop? Your existing client connects to port 6379 unmodified.

3. Watch it decide#

Open the dashboard and you'll see every verdict streaming live — hits by type, misses, safety blocks, latency, and an estimate of what the cache saved you:

shell

open http://127.0.0.1:6380

4. Wire it into your app#

The SDKs wrap the whole cache-or-compute loop in one call. If the answer is cached and safe to reuse, you never touch the model:

pip install crowkis

from crowkis import CrowkisClient

cache = CrowkisClient(host="127.0.0.1", port=6379, tenant="demo", model="gpt-4o")

answer = cache.get_or_compute(
    "Explain vector caches",
    lambda query: call_llm(query),
    ttl=3600,
)

Node shop? npm install crowkis and getOrCompute — same pattern. Using Claude Code or agents? See MCP for AI apps.