Models

Model abstraction layer (routing + fallback + eval hooks)

This is an API-first v1 model routing layer used by the orchestrator and pilot brief generator. Default mode is simulation for local demos; enable real calls via environment variables.

Orchestrator Observability

Routing configuration

Strategy

—

Primary model

—

Fallback model

—

Default tier

—

Fallback default

undefined

Allowed models

(all)

Set UMAMIMIND_LLM_MODE=openai and OPENAI_API_KEY to enable real model calls.

Registered models (v1)

Model

Tier

Provider

$/1k tok

No models found.

What this unlocks

Multi-model routing (explicit or tier-based) without touching product flows.
Fallback behavior you can govern via env vars (per tenant later).
Eval hooks recorded on LLM actions (PII/secret-like heuristics for demo evidence).
Clean integration point for future: eval-driven routing, vendor diversification, and cost/SLA policies.