Built for every workload

One engine. Every workload.

From AI assistants, agents and retrieval to caching, queues, leaderboards and real-time analytics. See each workload animated step by step, with the real commands, on one multi-threaded HotKV™ server.

use cases: 31
AI and LLM: 16
core data: 15
engine to run: 1

Step 1 of 6

Cache hit

A product page is served from memory. The database is not touched.

Example sessionplaying

# Cache hit

hotkv> GET product:1001

AI and LLM

16

LLM applications

AI customer support assistant

Answer repeats from cache, paraphrases by meaning and new questions from your help center, within a budget per customer.

PCACHE.GETSCACHE.GETRAG.SEARCH

LLM applications

Caching LLM gateway

Point your OpenAI or Anthropic SDK at one URL and stop paying twice for the same request. No code changes.

LLM applications

Semantic response caching

Cache LLM answers by meaning, so reworded questions hit the cache and skip the model call.

SCACHE.CREATESCACHE.GETSCACHE.SET

LLM applications

Exact prompt caching

Never pay twice for the same prompt, with hits, latency and money saved reported per namespace.

PCACHE.CREATEPCACHE.SETPCACHE.STATS

LLM applications

Prompt versioning and rollback

Version system prompts in the data layer and roll back in one command, with no redeploy.

PROMPT.PUTPROMPT.GETPROMPT.SETACTIVE

RAG knowledge assistants

Chunking, metadata filters, hybrid vector and keyword search and cached rerank scores in one engine.

RAG.CREATERAG.INGESTRAG.SEARCH

Vector similarity search

Find the nearest items by meaning, filtered by tag, next to the data the items describe.

VSIM.CREATEINDEXVSIM.SETVSIM.SEARCH

Personalization

Real-time recommendations

User features, similar items, live trends and seen-item filters, read from one server per request.

FEATURE.GETVSIM.SEARCHTOPK.LIST

Agents with memory

Conversation windows, semantic recall, working memory and an action log for long-running agents.

AMEM.CONV.APPENDAMEM.SEM.RECALLAMEM.WORK.SET

Multi-agent orchestration

Agents hand off work through a durable task stream and share memory and tool results.

XREADGROUPXAUTOCLAIMTOOL.CACHE.GET

Agent tool call caching

Memoize deterministic tool calls so agents stop paying for the same API call twice.

TOOL.CACHE.CREATETOOL.CACHE.GETTOOL.CACHE.SET

LangGraph checkpoints

Pause, resume and rewind LangGraph agents, with long-term memory searchable by meaning.

Memory for AI assistants over MCP

Give Claude, Cursor or your own agent notes, memory and document search through a safe MCP server.

LLM cost governance

Token and spend budgets per tenant, charged atomically before the model is called.

LLMBUDGET.SETLLMBUDGET.CONSUMELLMSTATS.RECORD

Guardrail and moderation caching

Replay answers that already passed validation and turn away repeated refused prompts at once.

GUARD.SETGUARD.GETNCACHE.SET

Machine learning

Real-time ML feature store

Typed features written by your pipelines and read by your models at inference time, with their age.

FEATURE.CREATEFEATURE.SETFEATURE.GET

Core data

15

Caching

Sub-millisecond reads, TTLs down to nanoseconds and invalidation by tag, on every core.

SETNEXPIRETAG.SET

Session management

Sessions that expire on their own and can be found or revoked by user, role or device.

TAG.SETTAG.MEMBERSTAG.INTER

Rate limiting

Atomic limits in one command, with GCRA, sliding window, token bucket or leaky bucket.

GCRA.LIMITRATELIMIT.CREATERATELIMIT.CHECK

Real-time leaderboards

Rankings that stay sorted as scores arrive, with reads and writes in parallel.

ZADDZINCRBYZREVRANGE

Background work

Job queues

FIFO queues that never lose a job when a worker crashes.

BLMOVEBRPOPLPUSH

Background work

Event streaming

Streams with consumer groups, acknowledgements and recovery of stuck messages.

XADDXREADGROUPXACK

Real-time chat and notifications

Fan messages out to every connected client, with sharded channels and history.

PUBLISHSUBSCRIBESPUBLISH

Distributed locks

Locks that expire on their own and can only be released by their owner.

SETDELEXDIGEST

Flash sales and inventory

Sell exactly the stock you have when thousands click Buy at once, and keep the bots out.

DECRSETGCRA.LIMIT

Real-time fraud detection

Velocity counters, blocklists, frequency estimates and model features in the payment path.

BF.EXISTSCMS.QUERYFEATURE.GET

Geospatial search

Find the nearest drivers, couriers or stores from live positions.

GEOADDGEOSEARCHGEODIST

IoT and metrics time series

Ingest samples, keep them as long as needed, downsample and aggregate on demand.

TS.CREATETS.ADDTS.DOWNSAMPLE

Real-time analytics

Count events as they happen, rank what is trending and spot hot keys.

INCRZINCRBYHOTKEYS

Counting at scale

Unique users, membership, top items and frequencies, each in a small, fixed amount of memory.

PFADDBF.EXISTSTOPK.LIST

Multi-tenant SaaS backends

Many customers on one cluster, isolated by ACL, with a registry of tenants and their users.

TENANT.CREATETENANT.BINDTENANT.STATS

By industry

Real systems, built from these pieces

Most products combine several workloads. Here are common combinations, each linked to the pieces it uses.

AI products and copilots

Assistants that cache, retrieve, remember and stay within budget.

E-commerce and marketplaces

Fast catalogs, honest stock counts and recommendations that convert.

Fintech and payments

Fraud signals in the payment path, strict limits and safe locking.

Gaming

Live rankings, player sessions, chat and job queues for match results.

Media and streaming

Personal home screens, trending content and real-time audience counts.

Mobility and delivery

Live positions, nearest-driver matching and dispatch without double booking.

IoT and industrial

Telemetry ingestion, downsampling and event processing for fleets of devices.

B2B SaaS platforms

Tenants on a shared cluster, per-plan limits and background jobs.

Your workload is not on the list?

HotKV speaks the Redis protocol and adds AI engines of its own, so the patterns you already use carry over.

Get started Read the docs