02/Model Gateway
Nodes
The physical LLM providers that serve requests.
A node is an LLM server the gateway talks to. Each node has a type, address, credentials, health and live metrics.
#Node types
| Type | Description |
|---|---|
| ollama | Ollama server (local/remote), native API. |
| openai | Any OpenAI-compatible provider (OpenAI, OpenRouter, Nvidia, Anthropic-compatible, +~37 catalog). |
| vllm | Self-hosted vLLM (OpenAI-compatible path). |
| bedrock | AWS Bedrock (specialized provider path). |
| cursor / antigravity | IDE-specific integrations (specialized provider path). |
#Health & metrics
For each node the platform tracks health checks, model sync and live metrics (avg response time, total requests, error count, active requests). The Statistics card on a node's detail page shows these; the same data feeds the score router.
Metric attribution works for all node types (ollama, openai/vllm, bedrock, cursor…). For openai/vllm nodes, base_url normalization ensures the metric is written to the correct node.