Questions, answered plainly.
On piqc, pricing, the founding customer program, and what happens to your data.
Getting started
Is piqc really free, forever?
Yes. piqc is open source and free to run against any Kubernetes cluster — a single command gets you a full GPU waste report, no account or payment required. It stays free regardless of which Paralleliq platform tier, if any, you adopt later.
How do I get started for free?
Install piqc — no signup, no credit card. It's a read-only scanner: point it at your cluster and it prints a cost report to your terminal. See the self-serve tier for three install options (Kubernetes Job, CLI, or Docker).
Do I have to buy the whole platform, or can I start with just detection?
You can adopt Detect on its own for continuous visibility with no remediation workflow, and add Decide & Fix later once you're ready for recommendations to route through an approval queue. Fleet Scale is additive on top of either, for teams managing multiple clusters — including customers' clusters.
Product
What makes Paralleliq different from standard infrastructure monitoring?
Standard infrastructure monitoring tells you GPU utilization. It does not tell you whether the model running on that GPU belongs on that hardware tier, whether the serving engine is configured to extract full throughput, whether an allocated node is generating any token revenue, or whether KV cache pressure is building toward an OOM event. These are model-aware signals that require understanding the relationship between the model, the hardware, and the serving configuration.
Does Paralleliq make changes automatically?
No. Every recommended action requires operator approval before anything touches the fleet. The optimization engine is rules-based and deterministic — not AI-driven. Each recommendation shows the blast radius and cost impact before you approve it. Every approved action is logged permanently under a named operator identity, creating an immutable audit trail.
Can reducing GPU waste lower my cost per token?
Yes — and it also expands effective capacity without procurement. When models run on the wrong hardware tier, when nodes sit allocated but serving zero tokens, or when serving engines are misconfigured below their throughput ceiling, every token costs more to produce than it should. Recovering that efficiency gives you a choice: improve margin, lower prices, or serve more customers from the same hardware.
Programs & pricing
Why don't you list prices?
Fleet size, GPU mix, cluster topology, and whether you need on-prem/air-gapped deployment all move the number enough that a single published price would be misleading more often than it would be helpful. Tell us about your fleet and we'll give you a real number quickly — not a multi-week sales process.
What's a founding customer, and how is that different from a design partner?
Same program, two names you may see used on the site. We're pre-launch and working with our first handful of customers directly — in exchange for early access, direct engineering collaboration, and input on the roadmap, founding customers get preferred pricing. Apply from the Guided Onboarding or Managed tier pages.
Trust & data
Do you have case studies or existing customers?
Not yet as a Paralleliq customer — we're pre-launch and working with our first design partners. The case studies on this site are real, though: they're results our own team members achieved in prior roles, at other companies, using different tools, on the same class of problem Paralleliq now solves directly. Each one is labeled to make that distinction clear. If you'd like to be among our first customers, see the founding customer program above.
Does my data ever leave my cluster?
No. Paralleliq deploys entirely within your environment — your telemetry, model weights, inference inputs, and workload data never leave it, and we have no access to them. piqc is read-only by design. Full detail on security, availability, and compliance is on the Trust & Security page.
Didn't find your question? Ask us directly.