AgentGateway Playbook

What's served, how to call it, and how routes map to backends

← Back to portal
You are not signed in. The model list below is fetched live with each user's own API key — sign in to see what's available to you specifically.

This cluster runs two separate AgentGateway deployments — same backend models, two different proxy technologies. Use whichever endpoint you prefer; your API key works against both (see the note on the standalone models list below if it doesn't for you).
  • AgentGateway (Kubernetes Gateway API) — https://agentgateway.dev.drai.auckland.ac.nz
  • AgentGateway Standalonehttps://agentgatewaystand.dev.drai.auckland.ac.nz

AgentGateway — models served

Sign in with an account holding the inference-consumer role and provision a key to see the live model list. Until then, here's the route inventory below.

AgentGateway — routes

MethodPathBackendNotes
POST /v1/chat/completions qwen38-27b (default) or gemma-4-26b Model-based routing on the same endpoint: send "model": "gemma-4-26B-A4B-it" for Gemma; any other value (e.g. "Qwen3.8-27B") routes to the primary Qwen chat model. There is no separate /v1/gemma/... path.
POST /v1/responses models-proxy OpenAI Responses API — routed by the model field, bypassing AgentGateway's Chat Completions schema validation (Responses uses "input", not "messages").
POST /v1/embeddings qwen3-embedding-4b (default) or f2llm-v2-4b Model-based routing: send "model": "F2LLM-v2-4B" for F2LLM; any other value (e.g. "Qwen3-Embedding-4B") routes to the primary embedding model. Not a 50/50 split — deterministic by model field.
GET /v1/models models-proxy Lists every model the gateway can route to. Bypasses LLM validation.
POST /v1/cloud/openai/* openai Direct passthrough to OpenAI (external). Not currently functional — no real API key is configured on this cluster.
POST /v1/cloud/claude/* anthropic Direct passthrough to Anthropic Claude (external). Not currently functional — no real API key is configured on this cluster.

AgentGateway — quick start (cURL)

List models

curl https://agentgateway.dev.drai.auckland.ac.nz/v1/models \ -H "Authorization: Bearer YOUR_API_KEY"

Chat completion — Qwen3.8-27B (default model)

curl https://agentgateway.dev.drai.auckland.ac.nz/v1/chat/completions \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "Qwen3.8-27B", "messages": [{"role": "user", "content": "Hello"}] }'

Chat completion — Gemma (same endpoint, different "model")

curl https://agentgateway.dev.drai.auckland.ac.nz/v1/chat/completions \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "gemma-4-26B-A4B-it", "messages": [{"role": "user", "content": "Hello"}] }'

Embeddings — Qwen3-Embedding-4B (default model)

curl https://agentgateway.dev.drai.auckland.ac.nz/v1/embeddings \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "Qwen3-Embedding-4B", "input": "Embed this text."}'

Embeddings — F2LLM-v2-4B

curl https://agentgateway.dev.drai.auckland.ac.nz/v1/embeddings \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "F2LLM-v2-4B", "input": "Embed this text."}'

AgentGateway Standalone — models served

Sign in with an account holding the inference-consumer role and provision a key to see the live model list.

AgentGateway Standalone — routes

MethodPathBackendNotes
POST /v1/chat/completions qwen38-27b (default) or gemma-4-26b Model-based routing: send "model": "gemma-4-26B-A4B-it" for Gemma; any other value (e.g. "Qwen3.8-27B") routes to the primary Qwen chat model. Confirmed live end-to-end during setup.
POST /v1/embeddings qwen3-embedding-4b (default) or f2llm-v2-4b Model-based routing: send "model": "F2LLM-v2-4B" for F2LLM; any other value (e.g. "Qwen3-Embedding-4B") routes to the primary embedding model. Not independently re-tested since initial setup — same mechanism as chat, above.
GET /v1/models (all models) Lists every model this instance can route to. Confirmed live during setup.

AgentGateway Standalone — quick start (cURL)

List models

curl https://agentgatewaystand.dev.drai.auckland.ac.nz/v1/models \ -H "Authorization: Bearer YOUR_API_KEY"

Chat completion — Qwen3.8-27B (default model)

curl https://agentgatewaystand.dev.drai.auckland.ac.nz/v1/chat/completions \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "Qwen3.8-27B", "messages": [{"role": "user", "content": "Hello"}] }'

Chat completion — Gemma (same endpoint, different "model")

curl https://agentgatewaystand.dev.drai.auckland.ac.nz/v1/chat/completions \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "gemma-4-26B-A4B-it", "messages": [{"role": "user", "content": "Hello"}] }'

Embeddings — Qwen3-Embedding-4B (default model)

curl https://agentgatewaystand.dev.drai.auckland.ac.nz/v1/embeddings \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "Qwen3-Embedding-4B", "input": "Embed this text."}'

Embeddings — F2LLM-v2-4B

curl https://agentgatewaystand.dev.drai.auckland.ac.nz/v1/embeddings \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "F2LLM-v2-4B", "input": "Embed this text."}'

MCP servers & agents

Neither AgentGateway instance exposes MCP servers or agents on this cluster yet. When those backends are added (see agentgateway/backends.yaml), they will appear here automatically.