NYR is onboarding design partners with distributed AI workloads.Design partner program open.Discuss a workload
NYRPlan a rollout

One runtime for
distributed AI.

Deploy and operate AI workloads across cloud regions, edge locations, and customer infrastructure without rebuilding your stack for every environment.

Review a workload See how placement works
Placement
Requirements driven
Operations
Health aware
Boundaries
Policy controlled
Environments
Cloud to edge
CUSTOMER INFRA

One control plane across distributed compute

01Cloud regions02Edge locations03Customer infrastructure04On-prem clusters

Distributed AI without environment-by-environment rebuilds

Place every workload deliberately.
Keep it operating.

AI workloads have different latency budgets, hardware needs, data boundaries, costs, and failure modes. NYR gives product teams one runtime and control plane for placing and operating them across approved compute.

Review a workload

Built for AI product and platform teams that need more placement control than one distant endpoint can provide.

WORKLOAD REQUIREMENT“Meet the latency, hardware, availability, cost, and policy target wherever this runs.”
WITHOUT NYR

One endpoint forces every environment into the same trade-offs.

  • LATENCYNetwork distance consumes the response budget.
  • HARDWAREPlacement ignores the runtime and accelerator fit.
  • RESILIENCEOne serving path becomes a shared dependency.
  • POLICYData and ownership constraints become custom deployment work.
OUTCOMEMore infrastructure work and less control over placement.
WITH NYRRECOMMENDED PATH

One control plane places each workload by its requirements.

  • LATENCYEligible compute is evaluated against the workload target.
  • HARDWAREPlacement matches declared runtime and capacity needs.
  • RESILIENCEHealth and fallback policy inform the serving path.
  • POLICYOnly approved environments enter the placement set.
OUTCOMEOne operating model across unlike environments.
01 · LATENCYShorter paths when the workload needs them
02 · HARDWARERuntime requirements matched to capacity
03 · POLICYPlacement limited to approved boundaries
04 · HEALTHRecovery driven by declared fallback rules

What NYR is

The runtime and control plane
for distributed AI.

NYR places and operates AI workloads across distributed compute. Placement evaluates availability, latency, hardware, cost, and policy, then the runtime reconciles each approved environment toward the declared state.

Review the architecture
EXAMPLE / placement requestILLUSTRATIVE
POST /v1/deploy{ "workload": "assistant-v4", "placement": "distributed", "latency_target": "<25ms", "availability": "multi-zone", "data_policy": "approved-zone", "hardware": ["gpu", "cpu"] }
01

Policy boundary

approved
02

Eligible targets

example: 3
03

Selected profile

edge-gpu-02
LATENCY TARGET <25 msFAILOVER POLICY definedDATA approved zone

Illustrative workflow. Values show requested configuration, not live production telemetry.

From workload intent to running state

Declare once.
Operate across environments.

01MODELLATENCYUPTIMEDATA

Declare the workload

Define the version, runtime, compute needs, latency target, data boundary, and recovery policy.

02PATTERNSITE ASITE BSITE C

Register environments

Connect approved cloud regions, edge locations, and customer infrastructure with clear capability profiles.

03AICUSTOMERNEAR EDGETARGETPRIMARYFALLBACK

Place and deploy

NYR selects eligible compute using availability, latency, hardware, cost, and policy requirements.

04ONESITE 01SITE 02SITE 03

Operate continuously

Keep desired state, health, rollout, failover, and recovery visible across every environment.

For AI product and platform teams

Keep your product.
Add one operating layer.

  • YOU KEEPModels, application logic, product behavior, and customer relationships.
  • YOU CONNECTApproved cloud, edge, or customer-owned compute.
  • YOU DECLAREWorkload requirements, versions, policies, and fallback behavior.
  • NYR OPERATESPlacement, deployment state, health, rollout, failover, and recovery.
See the example placement request

Why placement matters

AI workloads should not all run
in one distant cloud.

01Latency
CUSTOMERNEAR EDGETARGET

Shorten the request path.

Place eligible inference closer to the request when network distance would consume the product's latency budget.

02Privacy
LOCAL ZONEAI

Respect approved data boundaries.

Run eligible processing inside an approved environment when raw inputs should not travel to a centralized cloud.

03Availability
PRIMARYFALLBACK

Design for changing conditions.

Use health-aware placement and declared fallback policy across approved capacity when a serving path becomes unavailable.

04Environment fit
ONE AISITE 01SITE 02SITE 03

Match the workload to the environment.

Place by hardware, runtime, network, cost, ownership, and policy instead of treating every location as identical.

Placement intelligence

Decisions that stay
explainable.

NYR combines declared workload requirements with environment capabilities and runtime health. It records why a target was selected and adapts infrastructure placement as approved conditions change. NYR does not train on customer data or alter product behavior.

NYRplacement intelligence
01Environment constraints
02Workload requirements
03Runtime health
observe → place → adapt infrastructure

About NYR

Building the operating layer
for distributed AI.

NYR is an early-stage infrastructure company working with design partners on real workloads that need more control over where and how inference runs.

Current programOne workload. One approved environment. One concrete placement review.
Talk with Ronen NYR on LinkedIn

Review a distributed workload

Where should your AI workload run?

Bring the workload, candidate environments, latency target, hardware needs, data policy, and fallback requirements. We'll map the placement constraints and show where NYR fits.

Review my workload