Skip to content

Agent Mesh

667 B
Raw Blame
# Bare-Metal GPU Reasoning Sidecar Node (node-inference.yaml).
#
# Sits beside the self-hosted open-weights LLM server (Ollama / vLLM serving
# DeepSeek-R1 or Gemma 3) and publishes it to the mesh as inference://private-llm.
# Enforces an inbound Datalog check so only on-premises private nodes may
# consume GPU inference capacity.
version: "v1alpha1"
labels:
  residency: on-prem
  zone: gpu-pool
  dept: finance
attenuation:
  checks:
    - 'check if label("residency", "on-prem");'
services:
  - type: inference
    name: private-llm
    description: "Self-hosted DeepSeek-R1 / Gemma 3 reasoning engine on bare-metal GPU"
    target_url: "http://127.0.0.1:18892"