Find the right hardware for your deployment

This decision engine weighs platform fit across compute capability, video stream capacity, power envelope, thermal and environmental suitability, and ecosystem alignment. A platform receives a higher match confidence when its realistic deployment characteristics align with the workload rather than just matching a single benchmark number.

• Compute fit for the selected model type and stream count
• Power compatibility with the stated deployment budget
• Environmental suitability for fanless, industrial, mobile, or rack deployments
• Platform flexibility for single-model versus multi-model pipelines
• Alternative recommendations when the top choice is power-, cost-, or cooling-constrained

• Primary recommendation: the best-fit hardware platform for the selected constraints
• Match confidence: a percentage score representing how well the platform aligns with the full decision profile
• Alternatives: secondary options that remain viable for the same workload
• Deployment metrics: compute TOPS, power range, stream capacity, cooling assumptions, and estimated cost
• Machine-readable JSON: a structured result that can be copied, shared, or reused by downstream systems

Elided from a real API response (8× 1080p object detection, August 2026 pricing). Full contract: API & agent docs.

POST /api/v1/hardware-selector
{"inputs": {"model": "object_detection", "scenario": "multi_camera",
            "resolution": "1080p", "streams": 8,
            "power": "moderate", "env": "edge_server_rack"}}

{
  "success": true,
  "results": {
    "recommended_device": "Jetson AGX Orin 32GB",
    "recommended_device_id": "jetson_agx_orin_32gb",
    "compute_tops": 200,
    "power_range_w": "15–40W",
    "cost_usd": 1799,
    "confidence_label": "medium",
    "recommended_configuration": {
      "deployment": { "node_count": 1, "per_node_assignment": [8],
                      "deployment_summary": "Single-node deployment with 8 streams" },
      "economics":  { "cost_per_node_usd": 1799, "total_cost_usd": 1799,
                      "cost_completeness": "complete" },
      "confidence": { "label": "medium",
                      "reason": "The single-node deployment is feasible. Capacity is backed by a measured workload benchmark." }
    },
    "alternative_configurations": [
      { "device": { "name": "Jetson AGX Orin 64GB", "id": "jetson_agx_orin" },
        "recommendation_category": "higher_memory",
        "economics": { "total_cost_usd": 2999 },
        "comparison_to_winner": { "summary": "$1,200 more." } }
    ]
  },
  "summary": "Jetson AGX Orin 32GB is recommended as a single-node deployment for eight 1080p multi-camera streams. …"
}

When should I choose Jetson over Coral TPU?

Choose Jetson when you need broader model support, CUDA-based workflows, robotics stack compatibility, or more general-purpose vision flexibility. Choose Coral TPU when ultra-low power and TensorFlow Lite deployment are the main constraints.

Does higher stream count always require a larger platform?

Usually yes. More concurrent streams increase compute requirements, sustained thermal load, and memory bandwidth pressure, which is why stream count is a primary sizing input.

Is fanless deployment treated as a hard constraint?

Not strictly. Fanless narrows the field, but platforms with a documented passive-enclosure integration path remain eligible with sustained capacity derated for passive cooling — the recommendation warns you and names the enclosure class. Platforms with no fanless integration path are excluded.

Does this tool assume quantized models?

The recommendation logic assumes practical deployment fit by platform class and workload type. In real deployments, quantization strategy, framework support, and model conversion constraints should still be validated before final hardware purchase.

What does the Hardware Selector decide?

The Hardware Selector recommends the best-fit edge AI compute platform based on deployment scenario, model type, stream count, power budget, and installation environment.