Local AI infrastructure
for NAM research.
The private AI server of the Network Architectures and Management Research Group.
Public HTTPS inference. VPN for account management and Grafana. Use your existing project key.
Model inference stays on lab hardware; web search is available through the API.
Model catalogue
Models
lab and lab-short run on DGX Spark. lab-long, lab-large and the Claude compatibility alias share the Blackwell engine. lab-small and lab-tiny share a CPU-only model.
| Alias & backend | Input / output | 1 request | Prefill | Shared concurrency limit | Measured aggregate | Informal positioning |
|---|
Speech APIs
Audio services
Account & access
From invitation to first request.
Accounts are invitation-based. Connect to the VPN for account management; use public HTTPS for inference without a VPN.
Integration guide
Use the API you already know.
LiteLLM provides authenticated OpenAI and Anthropic-compatible endpoints. Use your assigned virtual key and one of the stable model aliases.
Public HTTPS is the default. Use https://asterix.patras5g.eu/v1 from any network. The landing page and guides are public. Account management and Grafana remain available only through the VPN admin domain.
Platform capabilities