LLM Provider Management
Metadata-driven with intelligent routing
LLM provider management is Neurova's model infrastructure: unified administration of many LLM vendors, metadata-driven, with automatic discovery, intelligent routing and user isolation.
Core capabilities
| Capability | Description | Status |
|---|---|---|
| Metadata | capability / series / priority / scope metadata | Implemented |
| Automatic discovery | A three-tier probe chain (metadata → model list → capability probe) | Implemented |
| Intelligent filtering | Multi-dimensional selection by capability / series / vendor | Implemented |
| Candidate merging | Idempotent merge logic | Implemented |
| OpenCode | An additional OpenCode-compatible vendor | Implemented |
| Async streaming | Provider methods made async | Implemented |
| User isolation | Scope-level isolation, with API keys managed per user | Implemented |
Intelligent routing
request type detection (CHAT / VISION / TTS / STT / IMAGE_GENERATION ...)
↓
capability matching → health check → priority ordering → response time optimization → weighted scoring
↓
best model selected
API
GET /api/v1/provider # provider list (with metadata)
POST /api/v1/provider # add a provider
POST /api/v1/provider/discover # automatic discovery
GET /api/v1/model # model management