HTTP API

Aegaeon exposes an OpenAI-style API for deployed models and listens on port 8000 by default. The running service publishes request schemas and defaults at /docs, /redoc, and /openapi.json. Authentication is configured at your gateway.

Endpoints

Method	Path	Purpose
GET	/health	Fixed health response; no deep actor checks
GET	/v1/models	Deployed model list
GET	/v1/models/{model_name:path}	Deployed model metadata; the path converter accepts names containing /
POST	/v1/models/deploy	Deploy CPU weight cache replicas. See Model Management
POST	/v1/models/undeploy	Unload selected or all replicas. See Model Management
POST	/v1/chat/completions	Chat, nonstreaming or SSE
POST	/v1/completions	Legacy Completion, nonstreaming or SSE
POST	/tokenize	Tokenize a plain prompt
POST	/detokenize	Decode token IDs
GET	/v1/aegaeon/runtime	Cluster runtime snapshot
GET	/v1/aegaeon/events	Runtime SSE
GET	/v1/aegaeon/metrics	60-second access metrics and rate-limit settings

Errors

Status	Meaning
400	Invalid model or parameter, or insufficient KV-cache capacity for a request
422	Invalid request schema, including unsupported generation fields
429	Access rate limit
503	No READY replica
507	Deployment failed and no READY replica remains
500	Server error; inspect service logs

A KV-cache capacity failure after streaming starts is returned as SSE data containing an error object, followed by data: [DONE].

Runtime interfaces are described in Monitoring.