> ## 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.

# Deploy and operate AI models across your infrastructure.

> Build or connect inference with durable operations, autoscaling, safe releases, and evidence you can inspect.

<div className="ic-hero not-prose">
  <div className="ic-kicker"><span className="ic-status-dot" /> Open-source inference control plane</div>

  <div className="ic-actions">
    <a className="ic-button ic-button-primary" href="/quickstart">Start the quickstart <span aria-hidden="true">→</span></a>
    <a className="ic-button ic-button-secondary" href="https://github.com/infercrane/infercrane">View on GitHub <span aria-hidden="true">↗</span></a>
  </div>

  <div className="ic-proof">
    <span>Build or connect</span>
    <span>Safe releases</span>
    <span>Explainable operations</span>
    <span>No Kubernetes required</span>
  </div>
</div>

```bash Deploy a model theme={"theme":{"light":"github-light-default","dark":"vesper"}}
infercrane deploy mistralai/Mistral-7B-Instruct-v0.3 \
  --name support-production

# The operation continues if this terminal disconnects.
infercrane status support-production --watch
```

```bash Send a streaming request theme={"theme":{"light":"github-light-default","dark":"vesper"}}
infercrane request support-production \
  --message "Summarize this incident." \
  --stream
```

Build a deployment from a model or custom OCI workload, or connect inference you already run.
InferCrane gives applications one OpenAI-compatible logical endpoint while durable control-plane
operations provision, reconcile, scale, and safely replace workers behind it. Providers and runtimes
integrate through explicit contracts; Kubernetes is not required.

<div className="ic-flow not-prose" aria-label="InferCrane request path">
  <div className="ic-node"><span className="ic-node-label">request</span><strong>OpenAI client</strong></div>
  <span className="ic-flow-arrow" aria-hidden="true">→</span>
  <div className="ic-node ic-node-accent"><span className="ic-node-label">stable identity</span><strong>endpoint</strong></div>
  <span className="ic-flow-arrow" aria-hidden="true">→</span>
  <div className="ic-node"><span className="ic-node-label">policy</span><strong>route generation</strong></div>
  <span className="ic-flow-arrow" aria-hidden="true">→</span>
  <div className="ic-node"><span className="ic-node-label"><i className="ic-dot" /> healthy</span><strong>runtime replicas</strong></div>
</div>

## The operational layer around inference runtimes

<CardGroup cols={3}>
  <Card title="Build and deploy" icon="server" href="/showcase/build-inference">
    Start from a model, recipe, or immutable custom OCI workload. Disconnecting the CLI does not abandon the operation.
  </Card>

  <Card title="Protect releases" icon="shield-check" href="/features/release-guard">
    Compare active and candidate revisions with deterministic, persisted Release Guard policy.
  </Card>

  <Card title="Understand behavior" icon="wave-pulse" href="/features/explanations">
    Explain scaling, rollout, degradation, and cold starts from recorded state and measurements.
  </Card>
</CardGroup>

## Start from where you are

<CardGroup cols={3}>
  <Card title="I have a model" icon="cube" href="/showcase/build-inference">
    Plan a qualified serving configuration and create a durable endpoint.
  </Card>

  <Card title="I have an endpoint" icon="plug" href="/showcase/connect-existing">
    Add health and operational evidence without migrating or transferring lifecycle ownership.
  </Card>

  <Card title="I am evaluating" icon="flask" href="/quickstart">
    Run the complete GPU-free control and request path locally in under five minutes.
  </Card>
</CardGroup>

## One product, two compute modes

<Tabs>
  <Tab title="Elastic">
    A registered elastic backend provisions runtime replicas. InferCrane reconciles desired capacity, routes only to ready workers, and fences scale-down before provider termination.

    ```bash theme={"theme":{"light":"github-light-default","dark":"vesper"}}
    infercrane deploy mistralai/Mistral-7B-Instruct-v0.3 \
      --cloud runpod \
      --gpu L40S \
      --min 1 \
      --max 4
    ```
  </Tab>

  <Tab title="Serverless">
    A provider-native serverless backend owns worker allocation and scale-to-zero. InferCrane preserves the logical endpoint, lifecycle evidence, cancellation, and cleanup.

    ```bash theme={"theme":{"light":"github-light-default","dark":"vesper"}}
    infercrane deploy mistralai/Mistral-7B-Instruct-v0.3 \
      --compute serverless \
      --cloud runpod \
      --gpu L40S \
      --max 4
    ```
  </Tab>
</Tabs>

<Note>
  InferCrane is provider- and runtime-extensible by design. Provider and runtime capabilities are
  declared and conformance-tested; each adapter still has an independent qualification status. vLLM,
  SGLang, and custom OCI have separate evidence states, as do RunPod and narrow AWS BYOC adapters.
  Consult [capability status](/project-status) before relying on a combination.
</Note>

## Evidence, not guesses

InferCrane records durable operation events, immutable model and revision identity, normalized request telemetry, benchmark reproduction metadata, and deterministic policy decisions. It does not fabricate provider pricing or timing boundaries that the provider does not expose.

<CardGroup cols={2}>
  <Card title="Five-minute local tour" icon="rocket" href="/quickstart">
    Run the GPU-free development stack and send your first OpenAI-compatible request.
  </Card>

  <Card title="Read the architecture" icon="diagram-project" href="/architecture/system">
    See the control-plane and data-plane boundaries behind durable operations.
  </Card>

  <Card title="See production patterns" icon="sparkles" href="/showcase/index">
    Deploy a model, connect LiteLLM, scope a sandbox, attach trained artifact lineage, and reject a bad release.
  </Card>
</CardGroup>