Server and Hardware

Server and Hardware

The customer is responsible for providing a Linux server. Use the following specifications for optimal performance:

Component	Requirement / Recommendation
Operating System	Ubuntu 22.04 or later
CPU	4 cores (AMD, x86)
RAM	8 GB
Storage	128 GB
Database	PGvector (bundled with the platform deployment)

Processing services

The server above runs the core Intric application. Document and audio processing run as separately-scaled worker services — PDF pre-processing, document ingestion, audio transcription, plus a message queue and sandboxed code runner — and need capacity in addition to the base server. PDF pre-processing dominates the footprint (each replica reserves 6 cores / 5 GB), so the total tracks how many PDF workers you run.

Two sizing tiers, by worker count per service. Reserved is the capacity that must be schedulable for the workers to run; under full burst, CPU usage can rise to roughly twice the reserved figure, so leave headroom.

Tier	PDF pre-processing	Document ingestion	Audio transcription	CPU (reserved)	RAM (reserved)
Minimum	4	4	3	~28 cores	~26 GB
Recommended	9	9	5	~60 cores	~57 GB

The message queue and supporting workers add a few more cores; PDF pre-processing is roughly 90% of the total.

The charts ship conservative replica ceilings — consumers.pdfPreProcessing.scaling.maxReplicas defaults to 2, and document ingestion to 5. Raise the ceiling for each worker to at least the count in your tier, or KEDA will not scale into the capacity you have provisioned.

Total capacity

Add the core application and operational headroom (including any self-hosted MCP servers):

Tier	CPU	RAM
Minimum	~40 cores	~50 GB
Recommended	~72 cores	~81 GB

(Core application 4 cores / 8 GB, plus processing services, plus 8 cores / 16 GB headroom.)

Network and Firewall

For Intric to function and be able to communicate with necessary services, the following network configurations are required.

Incoming traffic (Ports)

The following ports must be open and forwarded to the server:

• Port 22: SSH access (for operation and installation).
• Port 443: Web interface (HTTPS).
• Port 80: Automatic redirect to HTTPS.

Outgoing traffic (Whitelist - Infrastructure)

In addition to language models (see section below), the server needs access to the following services for infrastructure and code:

• github.com (Fetching application code)
• ghcr.io (GitHub Container Registry)
• *.docker.io (Docker Hub)
• login.intric.ai (Intric’s IdP is used for authentication)

Language models for on-prem operation

For Intric to function correctly, the system needs to communicate with external language models and AI services. The following addresses and IP numbers must be allowed for outgoing traffic in your firewall:

Intric Hosted Models (EU):

• 65.108.33.103 (Intric hosted Gemma 3 EU)
• 65.109.75.50 (Intric hosted Multilingual-E5-Large EU & Whisper)
• 77.87.121.4 (Intric hosted Gemma 3 SWE)
• mcp.intric.ai/* (Intric hosted MCP servers)
• mcp-servers.intric.ai/* (Intric hosted MCP servers)

External Public Models (APIs):

• api.openai.com/v1
• api.anthropic.com
• api.berget.ai/v1
• api.mistral.ai/v1

Domain and Certificate

For the application to be securely accessible for end users, the following is required:

1. DNS: Point your chosen domain (e.g., app.company.se) to the server’s IP with an A or AAAA record.

2. SSL/TLS certificate: You need to provide valid certificates:

   • Certificate file (.crt): Full chain in PEM format.
   • Private key (.key): Private key in PEM format.
   • Note: The customer is responsible for certificate renewal.