GitOps for AI Infrastructure

Declarative Blueprints & CI/CD Pipelines

The core philosophy of this architecture is GitOps for AI Infrastructure. Engineering teams do not need to manage disparate cloud accounts, install CUDA drivers, configure Kubernetes clusters, or maintain complex Terraform state files. Your private repository contains only configuration, and a simple CI/CD workflow triggers OneInfer to handle the heavy lifting.

Open Blueprints Catalog The Two-File Contract Under the Hood

GitOps Workflow Lifecycle

1Developer Commits

Developer commits oneinfer.yaml into private GitHub or Azure DevOps repo.

2CI/CD Triggers

git push triggers GitHub Action / ADO pipeline with X-API-Key.

3Zero-Touch Provision

OneInfer reserves GPU, pulls container via udocker, and binds intelligent router.

4Live AI Endpoint

Pipeline outputs live OpenAI-compatible URL & auth key ready for production traffic.

The Two-File Contract

To deploy any AI model or pipeline, you only need two files in your repository:

File A

The Configuration File: `oneinfer.yaml`

You define what you want, not how to provision the cloud:

version: "1"
name: "my-production-rag"

compute:
  type: "gpu"
  gpu_type: "RTX_4090"
  provider: "auto"             # OneInfer picks the lowest cost / highest availability
  container:
    image: "vllm/vllm-openai:latest"
    port: 8000
    command: "python3 -m vllm.entrypoints.openai.api_server --model deepseek-ai/DeepSeek-R1-Distill-Qwen-14B"

routing:
  enabled: true
  strategy: "adaptive"         # Routes to local GPU first, cloud fallback if overloaded
  primary_model: "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B"
  fallbacks: ["gpt-4o-mini"]
  middleware:
    semantic_caching: true     # Saves cost on repeated queries

File B

The Pipeline File: `.github/workflows/deploy.yml`

A standard ~15-line script that triggers deployment on git push:

name: Deploy to OneInfer
on:
  push:
    branches: [ main ]

jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Trigger OneInfer Provisioning
        run: |
          curl -X POST "https://api.oneinfer.com/v1/developer/${{ secrets.ONEINFER_DEVELOPER_ID }}/blueprints/deploy" \
            -H "X-API-Key: ${{ secrets.ONEINFER_API_KEY }}" \
            -H "Content-Type: application/json" \
            -d @oneinfer.yaml

Alternatively, use the developer CLI in your pipeline: pip install oneinfer-cli && oneinfer apply -f oneinfer.yaml.

Repository Secrets

In your private GitHub or Azure DevOps settings, add two secrets copied from your OneInfer Console:

ONEINFER_DEVELOPER_ID

Your scoped organization / account identifier.

ONEINFER_API_KEY

Your secret API key with deployment permissions.

What OneInfer Does When the Pipeline Runs

When your pipeline runs, OneInfer executes an atomic 5-stage orchestration sequence:

Validates Quota & Credentials

Checks your developer wallet balance and verifies the cryptographic signature of your ONEINFER_API_KEY.

Finds Best GPU Hardware

Queries our multi-cloud pool (Nebius, RunPod, Vast.ai, etc.) for the requested GPU at the lowest cost and best latency.

Boots Container Runtime

Starts your specified container (vLLM, SGLang, FastAPI, or custom image) securely via user-space udocker runtime.

Registers Intelligent Router

Automatically attaches load balancing, semantic caching, rate limiting, and fallback gateways (e.g. GPT-4o-mini).

Issues Live Production Endpoint

Returns a live, production OpenAI-compatible URL directly in your pipeline logs (e.g. https://api.oneinfer.com/v1/endpoints/ep_98a7bc/chat/completions).

What You Never Have to Worry About

Traditional GPU Deployment	OneInfer GitOps Pipeline
Create & maintain accounts with multiple GPU clouds	Single OneInfer API key across all GPU providers
Manage NVIDIA CUDA drivers, kernel modules, & AMI images	Managed automatically by OneInfer container engine
Open firewall ports, SSH keys, security groups, & VPNs	Managed securely through OneInfer proxy/router
Build custom fallback logic & queues when GPU instances crash	Automatic cloud fallbacks built into declarative config

1-Click Deploy & Open-Source Badges

In addition to automated CI/CD pipelines, OneInfer supports 1-click deployments from the web console and public GitHub README badges:

README Badge

[![Deploy to OneInfer](https://oneinfer.com/deploy-badge.svg)](https://oneinfer.com/deploy?repo=YOUR_GITHUB_REPO_URL)

Ready to deploy your first blueprint?

Visit the Console Template Hub to launch a production DeepSeek R1 model, voice gateway, or enterprise RAG in 1 click.

Launch from Console Templates Hub