Endpoints View

The Endpoints view provides a management interface for the LLM endpoint registry.

Endpoint Table

Each configured endpoint is displayed with:

• Name -- Endpoint identifier from config.yaml
• Provider -- Adapter type (openai, anthropic, google, azure, ollama, etc.)
• Status -- Live (healthy), offline (failed health check), or discovered (auto-detected)
• Models -- Number of models served by this endpoint
• Latency -- Current EMA-weighted response latency
• Circuit -- Circuit breaker state (closed = healthy, open = failing)

Actions

Action	Description
Test	Run a model-listing probe and refresh status/latency/model metadata
Toggle	Enable or disable an endpoint without removing it
Delete	Remove an endpoint from the registry
Priority	Set routing priority for endpoint selection

Circuit Breaker

Endpoints that fail repeatedly are automatically circuit-broken:

• Closed -- Endpoint is healthy, accepting traffic
• Open -- Endpoint has failed, traffic is routed to fallback chain
• Half-Open -- Testing recovery with a single probe request

Circuit state is visible per-endpoint and triggers circuit_open / endpoint_recovered webhook events.

API

# Full registry state
curl http://localhost:8090/api/v1/registry \
  -H "Authorization: Bearer your-key"

# Toggle endpoint
curl -X POST http://localhost:8090/api/v1/registry/openai/toggle \
  -H "Authorization: Bearer your-key"

# Test endpoint without inference
curl -X POST http://localhost:8090/api/v1/registry/openai/probe \
  -H "Authorization: Bearer your-key"

# Set priority
curl -X POST http://localhost:8090/api/v1/registry/openai/priority \
  -H "Authorization: Bearer your-key" \
  -d '{"priority": 1}'