Adaptive Model Routing

Model Switchboard

Switchboard is 121's provider routing layer: a Foundation Beta path for choosing the right inference provider without hiding the state of credentials, cost limits, or live availability.

Early accessOpen API docs Admin status

Provider status

3 of 4 provider routes have required env vars configured.

Early access

Routing mode

Current mode: cost-optimized. Chain order applies mode policy, provider selection, context prefilter, diversity, then execution.

Early access

Credential boundary

Status exposes env var names and booleans only. No API key values are read into UI or responses.

Early access

Cost and quotas

Runtime quota counters and burn-rate summaries are active; Supabase persistence migration is staged.

Early access

Provider	Model	Status	Expected env vars
Cerebras	gpt-oss-120b	configured	CEREBRAS_API_KEY
Groq	llama-3.3-70b-versatile	configured	GROQ_API_KEY
OpenRouter	openrouter/free	configured	OPENROUTER_API_KEY
Ollama Cloud	minimax-m3:cloud	missing	OLLAMA_CLOUD_API_KEY, DODGER_OLLAMA_CLOUD_BASE_URL

GET /api/switchboard/status

Truth state: Foundation Beta
Routing mode: cost-optimized
Provider credentials configured: 3
Provider calls from status: false
Secret values exposed: false