> ## Documentation Index
> Fetch the complete documentation index at: https://infercrane.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Frequently asked questions

> Product scope, compatibility, data handling, serverless, sessions, and pricing answers.

# FAQ

## Does InferCrane support any model, engine, or cloud?

No. The current registry contains vLLM, SGLang, and immutable custom OCI profiles across specific
RunPod, AWS EC2 BYOC, GCP Compute BYOC, and namespaced Kubernetes modes. `infercrane integrations` reports the exact
qualification state; real-GPU evidence remains deferred until real-infrastructure qualification. Every exact
provider/runtime/mode combination needs independent evidence.

## Do I need Kubernetes?

No. The primary path does not require Kubernetes. A namespace-scoped Kubernetes adapter is available
for teams that already operate a cluster; InferCrane does not install a distribution or custom
operator.

## Does Release Guard use an LLM?

No. It applies a persisted deterministic policy to measured evidence.

## Does serverless mean InferCrane schedules GPUs?

No. The registered serverless provider owns worker allocation and scale-to-zero; InferCrane owns
logical lifecycle and evidence. RunPod Serverless is the first and currently only registered native
Serverless backend.

## Is InferCrane coupled to RunPod or vLLM?

No at the lifecycle boundary. Providers, runtimes, serverless status, artifacts, and benchmark tools
are external adapters composed around durable InferCrane state machines. `infercrane integrations`
shows the current exact combinations and separates simulated, local, deferred and real evidence.

## Are prompts or outputs stored?

Not by default. Telemetry and benchmark history contain measurements and operational metadata.

## Does InferCrane preserve agent or inference sessions?

Context Passport preserves bounded logical session identity and a preferred-backend hint. Reliability
always overrides affinity: when a worker disappears, the stale hint is removed and the next request
falls back to an ordinary healthy route. InferCrane does not store conversation bodies by default and
does not claim durable KV state or transparent request migration unless the selected backend declares
and qualifies that capability.

## Is provider pricing estimated?

Only trustworthy observed cost metadata is shown. InferCrane does not fabricate prices.