4 vCPU · 8 GiB RAM · 40 GiB workspace. Billed by the second, with $0 standby compute and no base subscription.
Create a managed sandbox →Launch pricing
Pay for measured progress, not platform promises.
Start with one bounded workload. We record the current route, test plausible execution paths, and hand back a qualified decision you can operate.
Compare engagement paths ↓Managed agent sandboxes.
Top up once. Pay only while compute runs.
Sleeping preserves files and stops processes. The minimum prepaid top-up is $10. The open-source runtime remains available on GitHub for teams that want to operate it.
How the bill works
Know exactly what creates the bill.
Open-source InferCrane in your environment.
A fixed, scoped workload qualification sprint.
Your BYOC cloud, GPU, storage, and model API charges remain yours.
Qualification and software fees stay separate from provider compute. No experiment or infrastructure spend begins before you approve the scope and budget.
Choose your operating model
One product. Three ways to adopt it.
Open SourcePublic launch
Freeopen-source software
Run InferCrane in your environment and keep full control of the stack.
- CLI, API, SDKs, and Terraform workflows
- Stable endpoints and existing-workload adoption
- Durable deployment, scaling, and recovery
- Request evidence and guarded releases
- Your cloud, cluster, runtimes, and provider accounts
Qualification sprint5 founding slots
€1,500from · plus compute
Turn one real workload into a measured execution decision and launch plan.
- Five-business-day scoped engagement
- Current route retained as the control
- Bounded model, runtime, configuration, and hardware search
- Quality, latency, throughput, reliability, and cost gates
- Qualified result, rejected candidates, and launch runbook
Managed operationsAfter qualification
€299from / month · plus compute
Keep the qualified path behind a stable endpoint while workload and supply evolve.
- Stable OpenAI-compatible endpoint
- Monitoring, spend limits, and operational evidence
- Release Guard approval and rollback
- Scheduled workload requalification
- Customer-owned or separately metered infrastructure
Enterprise deployments
Bring the boundary we need to prove.
Enterprise work starts with an exact infrastructure and governance requirement—not a generic SLA promise. We qualify the path together before commercial terms or production commitments.
Scope an enterprise deployment ↓- 01Private control plane
Self-hosted or BYOC networking, identity, and data boundaries.
- 02Exact qualification
A specific model, runtime, GPU, provider, and region tuple to prove.
- 03Controlled releases
Persisted evidence, audit history, approval, and rollback requirements.
- 04Operational ownership
Backup, restore, upgrades, support, and provider-failure expectations.
Common questions
Know the boundaries before you join.
01How are sandboxes priced?
Managed sandboxes cost $0.30 per active hour for 4 vCPU, 8 GiB RAM, and a 40 GiB workspace. Active compute is metered by the second. Sleeping compute is $0 and the workspace persists. There is no subscription.
02Does a sandbox keep cron jobs and services running while asleep?
No. Standby preserves the workspace but stops compute and processes. Choose resident compute for cron, daemons, and continuously reachable services. Resident compute consumes reserved capacity until it stops or expires.
03What does the qualification sprint cost?
Founding workload sprints start at €1,500 plus provider compute. We confirm the exact workload, evidence gates, timeline, infrastructure budget, and fixed engagement price before any work or spend begins.
04What happens after the sprint?
You receive the measured baseline, qualified winner, rejected candidates, evidence boundary, and launch runbook. You can operate it yourself, deploy it with InferCrane, or continue with managed operations from €299 per month plus infrastructure.
05Does InferCrane add a per-token fee to my own infrastructure?
No. BYOC infrastructure and external model-provider charges remain between you and those providers. The hosted product will use subscription pricing instead of adding a token markup to infrastructure you already pay for.
06Does InferCrane provide GPUs?
InferCrane can operate compute in your AWS, GCP, Kubernetes, or supported provider account, or qualify inference you already run. When InferCrane-managed capacity is available for the selected path, it is quoted and metered separately rather than hidden in the software fee.
07Can I self-host InferCrane?
Yes. The open-source distribution is intended to run in your environment. Enterprise deployments can include scoped architecture, qualification, and operational support for private infrastructure.
08Do I need to migrate my existing vLLM, SGLang, or gateway setup?
No. You can begin by connecting a compatible endpoint in observe-only mode. InferCrane can earn additional traffic and lifecycle ownership later without requiring a day-one replacement.
09Which models can I run?
InferCrane can work with models supported by the selected runtime or external endpoint. Every exact artifact, runtime, GPU, and workload combination still needs qualification before performance or reliability claims are trusted.
10What happens to prompts, credentials, and request data?
Credentials remain server-side and ordinary telemetry does not retain prompt or response content by default. Your chosen infrastructure and model providers still have their own data-handling policies, which remain part of the deployment decision.
11When do production SLAs apply?
Production SLAs and support levels apply only after the exact infrastructure path passes qualification and both parties agree to managed-operation terms. A qualification sprint alone does not create a production SLA.
Founding workload sprint
Bring one real workload. Leave with measured evidence.
Five business days from an agreed workload brief. Starting at €1,500 plus approved compute. Submitting this form creates an inquiry only—no charge or infrastructure.