SOP-002 · REV 1.0 · 2026-07-23

AI Deployment

Technical SOP for deploying private AI infrastructure on client hardware or isolated tenants.

Owner: Engineering Review Cycle: Monthly Classification: Internal

Pre-Deployment Requirements

Phase 1: Environment Setup

1

Node Provisioning

Prepare the target machine according to the deployment specifications.

  • OS: macOS 14+ or Ubuntu 22.04 LTS
  • RAM: 32GB minimum, 64GB recommended
  • Storage: 500GB SSD minimum
  • Network: Tailscale tailnet access configured
curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up --authkey=tskey-auth-...
2

LiteLLM Proxy Installation

Install and configure the model routing proxy.

pip install litellm
litellm --config /path/to/config.yaml --port 4000
Security: Ensure proxy is bound to localhost only (127.0.0.1:4000). Never expose directly to internet.
3

Memory Stack Setup

Deploy Qdrant vector store and Redis cache.

docker run -p 6333:6333 -v $(pwd)/qdrant_storage:/qdrant/storage qdrant/qdrant
docker run -p 6379:6379 redis:latest

Phase 2: Model Configuration

4

Ollama Installation

Install Ollama for local model inference.

curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3.1:70b
ollama pull nomic-embed-text
5

LiteLLM Config

Create configuration file routing to local Ollama.

model_list:
  - model_name: ollama/llama3.1:70b
    litellm_params:
      model: ollama/llama3.1:70b
      api_base: http://localhost:11434
  - model_name: ollama/nomic-embed-text
    litellm_params:
      model: ollama/nomic-embed-text
      api_base: http://localhost:11434

Phase 3: Security & Compliance

6

Data Isolation Verification

7

Anonymization Pipeline Test

Verify PII detection and redaction before storage.

python -m oction.deid.test --client-id CLIENT_ID --sample-data /path/to/test/docs
All client data must pass through the de-identification pipeline before vector storage.

Phase 4: Handoff

8

Health Check

curl http://localhost:4000/health
./oction-node.sh status
9

Documentation