Prerequisites

Hardware Requirements

Profile	RAM	CPU	Disk	Use Case
Development	4 GB	2 cores	20 GB	Local dev, testing
Small team	8 GB	4 cores	50 GB	1-10 users, low volume
Production	16-32 GB	8+ cores	100+ GB SSD	10-100+ users, high volume

RAM allocation

PostgreSQL with pgvector and the Celery worker are the most memory-intensive components. For production, allocate at least 4 GB to PostgreSQL and 2 GB per Celery worker concurrency slot.

Disk Considerations

• PostgreSQL data grows with document count and embedding dimensions (1536-dim vectors at ~6 KB per chunk)
• Uploaded files are stored in the cl-storage volume — plan for your expected document sizes (the default upload limit is 100 MB per file)
• Redis memory usage is minimal (task queue only), typically under 100 MB

Software Requirements

Container Runtime (recommended)

Software	Minimum Version	Notes
Docker Engine	24.0+	Or Docker Desktop 4.20+
Docker Compose	v2.20+	Included with Docker Desktop

Kubernetes (alternative)

Software	Minimum Version	Notes
Kubernetes	1.28+	Any distribution (EKS, AKS, GKE, k3s)
Helm	3.12+	For chart-based deployment
kubectl	Matching cluster version	For management

Without Containers

If running services directly on a host:

Software	Version
Python	3.12+
Node.js	20 LTS+
PostgreSQL	16+ with pgvector extension
Redis	7+
Tesseract OCR	5+ (with English language pack)

Network Requirements

Outbound (from your server)

Destination	Port	Purpose	Required?
api.anthropic.com	443 (HTTPS)	Anthropic Claude API	If using Anthropic
api.openai.com	443 (HTTPS)	OpenAI API	If using OpenAI
Docker Hub / GHCR	443 (HTTPS)	Pull container images	Initial setup and updates

At least one AI provider is required

Contract Lucidity requires at least one AI provider (Anthropic or OpenAI) to function. The document processing pipeline will not start without a configured provider. You can configure providers after first boot via Settings > AI Capabilities.

Inbound (to your server)

Port	Service	Access
443 (HTTPS)	Reverse proxy / load balancer	All users
3000	cl-frontend (if no reverse proxy)	All users (development only)

Production deployments

Never expose ports 5432 (PostgreSQL), 6379 (Redis), or 8000 (backend API) to the public internet. These should only be accessible within the Docker network or private subnet.

AI Provider Account

You need an API key from at least one of:

Provider	Sign-up	Models Used
Anthropic	console.anthropic.com	Claude (Haiku, Sonnet, Opus) for analysis; no native embedding model — pair with OpenAI for embeddings
OpenAI	platform.openai.com	GPT-5.4 nano/mini/full for analysis; text-embedding-3-small for embeddings

Contract Lucidity maps AI capabilities independently, so you can mix providers — for example, use Anthropic Claude for document analysis and OpenAI for embeddings.

AI Capabilities

The system uses five AI capability slots, each assignable to a specific provider and model:

Capability	Used For
extraction_classification	Document type classification, metadata extraction
document_understanding	Clause extraction, structured data extraction, report generation
embeddings	Vector embeddings for semantic search and playbook matching
reasoning	Complex analysis tasks
generation	Content generation (drafts, counter-proposals)

Domain and SSL

For production deployments:

• A domain name (e.g., contractlucidity.yourcompany.com)
• A valid SSL/TLS certificate (Let's Encrypt, corporate CA, or cloud-managed)
• A reverse proxy (Nginx, Caddy, Traefik, or cloud load balancer) terminating TLS

Storage

Environment	Recommendation
Development	Local Docker volumes (default)
Small team	Local disk or NAS mount
Production	Azure Files, AWS EFS/S3, GCP Filestore, or NFS

The backend writes uploaded documents to /data/storage and configuration data to /data/config. Both paths are Docker volumes (cl-storage and cl-config) that must be backed up.