Helm Inference API

Raw MetalMind inference endpoint. Direct access to PhotonicGPT-v2 token generation. Sovereign — no external APIs.

Endpoint

• POST /v1/infer — raw inference

Raw Inference

POST /v1/infer
Content-Type: application/json

{
  "prompt": "The sovereign field computes",
  "tokens": 100,
  "temperature": 0.85,
  "topk": 30
}

Response:
{
  "output": "..."
}

Parameters

• prompt — input text (required)
• tokens — max tokens to generate (default: 200, max: 400)
• temperature — sampling temperature (default: 0.85)
• topk — top-k sampling (default: 30)

Stack

• flowistan_tls (443) → weyland :7720 → metalmind_cpu → detok.sh
• Weights: photonic_mind.RAW (PhotonicGPT-v2, 14M params)
• Node: g0 (5.161.253.15)