> ## 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.
# Choose the right inference approach
> Compare InferCrane with managed inference platforms, model APIs, gateways, and Kubernetes-native stacks.
# Not every team needs an inference control plane
Choose based on the responsibility you want to own. InferCrane is designed for teams that want
control of their infrastructure or existing inference stack without building the operational layer
from scratch.
| Approach | Best when | You delegate | You retain |
| -------------------------- | ------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------- | ------------------------------------------------------------------------------- |
| Managed inference platform | Fastest hosted path matters more than infrastructure portability | Compute, scheduling, builds, scaling, and much of operations | Model/application behavior |
| Hosted model API | You want an API call and no model-serving operations | Model weights, runtime, infrastructure, and capacity | Application and provider selection |
| AI gateway | You need one API across managed model providers | Protocol translation, provider routing, and accounting | Provider accounts and workload deployment |
| Kubernetes-native serving | Your platform team already operates Kubernetes and wants native resources | Cluster reconciliation primitives | Cluster, networking, storage, GPU scheduling, and serving stack |
| InferCrane | You want a stable endpoint and evidence-driven lifecycle across infrastructure you control | Durable operations, revisions, rollout policy, explanations, and adapter contracts | Provider accounts, runtime choice, infrastructure visibility, and data boundary |
## InferCrane and managed platforms
[Baseten](https://docs.baseten.co/overview) and [Modal](https://modal.com/docs/guide) offer managed
compute and highly integrated hosted experiences. They are usually a better fit when the team wants
the vendor to own infrastructure and capacity from day one.
InferCrane is differentiated by customer-controlled infrastructure, incremental adoption of existing
endpoints, replaceable provider/runtime contracts, persisted operational decisions, and explicit
qualification boundaries. It is not currently a managed-compute substitute.
## InferCrane and gateways
LiteLLM and OpenRouter solve different layers:
* A user-managed LiteLLM gateway can translate and route among managed model APIs.
* OpenRouter can provide external model capacity through its hosted API.
* InferCrane manages stable endpoint identity, lifecycle state, revisions, infrastructure-backed
deployments, rollout evidence, and bounded external overflow.
These tools can compose. InferCrane can connect a LiteLLM endpoint or use explicitly governed
OpenRouter capacity; it does not fork or silently replace either product.
## InferCrane and Kubernetes inference stacks
KServe, llm-d, NVIDIA Dynamo, and Kubernetes Gateway API provide powerful cluster-native serving and
data-plane capabilities. InferCrane can use qualified infrastructure/runtime integrations while
remaining the user-facing lifecycle and evidence layer. It does not build another Kubernetes
operator or reimplement distributed serving.
## The decision shortcut
You want the vendor to own compute and the fastest route to a hosted endpoint.
You want to consume hosted models and do not need to operate weights or runtimes.
You already have a platform team and want cluster-native serving resources.
You want to build or connect inference on your infrastructure with durable operations and evidence.
This comparison describes product categories, not a benchmark. Verify current vendor capabilities,
pricing, support, and deployment boundaries directly with each provider.