TypeSafe Jev (System One Decision Model)
TypeSafe Jev is a System One decision model delivering sub-100ms latency, zero-token generation overhead, typed scoring, and no charge for output tokens.
typesafe/jev is a System One decision model: a non-generative engine for classification, intent routing, policy verification, and structured decision scoring. Where a generative LLM answers token by token, Jev scores the input in a single forward pass and returns a typed decision — sub-100ms latency, zero-token generation overhead, and no charge for output tokens.
| Attribute | System One: TypeSafe Jev (typesafe/jev) | System Two: Generative LLMs |
|---|---|---|
| Primary Function | Fast, reflexive classification, intent routing, policy enforcement | Deep multi-step reasoning, creative synthesis, open-ended dialog |
| Execution Model | Single forward pass probabilistic decision scoring | Autoregressive token-by-token sequence generation |
| Typical Latency | Sub-100ms | Hundreds of milliseconds to seconds |
| Output Token Risk | None (zero generation overhead) | High variance based on output token volume |
Billing & Auto Routing
- Input tokens: Billed at flat provider list price with zero markup; output tokens are not charged.
- Context: 128,000 tokens input, 1,024 max decision output.
nrouter/auto: Powers Best for the task under a 2,000 ms decision deadline (configurable to 5,000 ms). If it exceeds the deadline, the request falls back to the tier default.- Privacy: Disable Smart model selection in Settings → Privacy to prevent sending prompt text to Jev.
Direct API Call
Call typesafe/jev via /v1/chat/completions:
curl https://api.nrouter.ai/v1/chat/completions \
-H "Authorization: Bearer $NROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "typesafe/jev",
"messages": [
{ "role": "system", "content": "Classify user intent: [BILLING, TECHNICAL, SALES, CHITCHAT]." },
{ "role": "user", "content": "How do I upgrade to Enterprise?" }
],
"response_format": { "type": "json_object" }
}'Returns a typed JSON decision with calibrated confidence score (0.0 to 1.0).
Frequently asked questions
What happens if TypeSafe Jev exceeds its deadline during an nrouter/auto call?
The request falls back automatically to the tier's reviewed default model without refusal.
Can our organization use nrouter/auto without sending prompt text to TypeSafe Jev?
Yes; disable Smart model selection under Settings → Privacy and select Lowest cost in Router Settings.
Why are extra_content, thought_signature, and system_fingerprint stripped on nrouter/auto?
Upstream vendor fingerprints and signatures are masked to maintain a clean, provider-agnostic response interface.