VCAL Project
On-prem / VPC Model-agnostic Agentic workflows Prometheus + Grafana Open core

VCAL Semantic Cache

Put VCAL Semantic Cache server in front of your LLM application, RAG system, or agentic workflow. Repeated or similar prompts get answered from your private cache with millisecond latency. New prompts go to the model as usual — and VCAL Semantic Cache server learns them for next time.

Typical savings
30–60%
Latency on hits
Milliseconds
Data control
Your perimeter
Time to value
Days

Useful for repeated questions, RAG answers, support assistants, and agent steps such as routing, summarization, validation, and reuse.
Core engine is open source: vcal-core. Production service is commercial: VCAL Semantic Cache.

Open core + commercial server

VCAL follows an open-core model: the indexing engine is public and auditable, while the production server ships as signed binaries/containers for LLM applications, RAG systems, and agentic workflows.

vcal-core (Open Source)

High-performance Rust library for semantic caching. Embeddable and auditable.

  • • HNSW similarity search
  • • Snapshots • TTL • LRU eviction
  • • Used by VCAL Semantic Cache
View vcal-core on GitHub →

VCAL Semantic Cache (Commercial)

Production-ready semantic cache service built on vcal-core. Adds observability, licensing, and support.

  • • HTTP API + OpenAPI
  • • Prometheus metrics + Grafana dashboards
  • • On-prem / VPC deployment
Explore VCAL Semantic Cache →

VCAL Semantic Cache

VCAL Semantic Cache is the production HTTP service built on vcal-core. It adds licensing, observability, and deployable artifacts for teams that need semantic reuse across LLM applications, RAG systems, and agentic workflows.

Integrating VCAL into your LLM or agentic pipeline

VCAL integrates as an HTTP semantic cache that your application or agent queries before calling the LLM. If a similar question, answer, or agent step was handled before, VCAL returns the cached answer immediately. Otherwise, your application calls the model and stores the result for future reuse.

VCAL integrates as a lightweight HTTP semantic cache in front of any LLM provider, application backend, or agent framework.

View integration examples →

Why teams deploy VCAL

Reduce paid model calls, speed up repeated LLM and agentic workflows, keep answers inside your perimeter, and prove the ROI in dashboards.

Cut repeat model calls

Serve repeats from cache instead of paying your LLM again and again.

Millisecond hits

Lower tail latency on repeated questions — users feel the difference instantly.

Agent workflow reuse

Reuse repeated outputs from routing, summarization, validation, support, and internal assistant flows.

Private by design

Run on-prem or in VPC. With VCAL Semantic Cache: metrics, snapshots, auth, and enterprise options.

Benchmarks

Fast cache lookups

  • p50: ~88 µs
  • p95: ~244 µs
  • QPS (single node): 27 K queries/s
  • Memory footprint: ~8 GB per 1 M vectors

ROI you can show

Use the ROI calculator below. Jump to ROI.

Sample metrics (Grafana — last 6 hours)

Grafana dashboard showing VCAL performance over the last 6 hours

Cache hit ratio, tokens saved, cost savings, answers cached, and requests number from a single-node VCAL Semantic Cache.

ROI calculator

Estimate monthly savings assuming cache hits skip paid LLM calls in applications, RAG systems, or agentic workflows.

Estimated monthly savings: $1,600

Assumes VCAL hits skip the paid LLM entirely (typical for repeat questions).

VCAL Semantic Cache Pricing

A 30-day evaluation is available through the CLI. Production licenses start from €9,900/year per environment. Enterprise deployments, security review, and MSP/reseller packaging are available by discussion.

View pricing Contact VCAL

MSP, reseller, and integration partners

Qualified MSPs, VARs, and integration partners can offer VCAL Semantic Cache as a managed AI cost-control service. Partner pricing, deployment support, and managed-service packaging are available on request.

VCAL Privacy Guard is available now. Security, Audit, and Compliance modules are planned for the broader VCAL Enterprise Guard stack.

Discuss partner pricing →

FAQ

Is this hard to roll out?

No. Run VCAL Semantic Cache in your environment, point your application to it first, and measure hit rate, latency, and savings within days.

Will our data leave our environment?

No. VCAL Semantic Cache runs on-prem or in your VPC. Your perimeter, your data.

Does VCAL work with our current model?

Yes. VCAL Semantic Cache is model-agnostic. Keep using OpenAI, Anthropic, Ollama, Hugging Face, local models, or other providers.

Do we need to use a specific embedding model?

No. Your application creates embeddings with the model or provider you choose, then sends vectors to VCAL for semantic lookup and reuse.

Can VCAL be used in agentic workflows?

Yes. VCAL is useful when agents repeat similar prompts or outputs, such as routing, summarization, validation, internal Q&A, support responses, or reusable reasoning steps. Your application decides when to query VCAL and when to call the model.

What’s open source vs commercial?

The core engine, vcal-core, is open source. VCAL Semantic Cache is the production service with licensing, observability, signed releases, deployment artifacts, and support.

Contact VCAL

Need pricing, procurement, or pilot details?

Contact VCAL for annual licensing, Privacy Guard pilots, enterprise deployment, security review, or MSP/reseller packaging.