JevCodeSystem One · A typed decision layer
Give your software deterministic AI decisions
Jev is not a chat model. It takes unstructured state plus a typed question and returns a typed decision — a choice, a score, or a boolean, each with a confidence. That makes it a drop-in decision layer for software.
What's new
Recently updated docs, community benchmarks, and projects just added.
Recently updated docs
View allCommunity pulse
View all- Diogo Almeida: After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not le…
- tamara: found the perfect use case for @typesafeai Jev: instant compaction in 2026, why is co…
- Matija Sosic: Here's a 45-second TL;DR on Jev. I find the core idea beautifully simple, but the video…
Three question primitives
Every decision reduces to one of three typed questions. Choosing the right primitive is the first step toward a reliable decision system.
Pick one from a set
Given a set of mutually exclusive options, the model returns the selected option with a confidence. Used for intent classification, ticket routing, action selection.
ScoreRate along a scale
Rate along a defined scale or rubric, returning a numeric value with confidence. Used for relevance ranking, quality assessment, risk grading.
NoulProbability of yes/no
Ask a yes/no question and get the probability the answer is yes. Used for verification, claim checking, agent guardrails.
Architecture patterns
From single calls to multi-stage pipelines, these patterns cover the decision architectures that show up most in production.
Intent routing
Dispatch requests to downstream handlers with a single Choice call, replacing brittle regex and keyword matching.
Confidence routing
Route by confidence: auto-execute when high, escalate to a human or a stronger model when low.
Composite scoring
Combine several Score dimensions into a single ranking metric, making ranking logic explainable and tunable.
Fan-out
Issue many independent judgements in parallel and aggregate, cutting total latency and raising throughput.
Ecosystem and real cases
Teams are already using Jev in production for classification and routing, verification, scoring and ranking, and agent guardrails.
Docs
Concepts, primitives, patterns, and SDKs, organized by task.
Quick start
A minimal call you can run in minutes.
Ecosystem
Community projects by category, with stars and the Jev decision point.
Community
Public benchmarks and discussion, each with its source.
40 documents available in this language.
