A pure-Rust port of Laya for typed decisions from your Rust code: give it a message and a few questions, and it answers each with a label, a score or a yes/no probability you can branch on. It helps with triaging support tickets, sorting emails, moderating posts and guarding prompts.

It is for cases where you need a reliable decision rather than generated text: classifying, scoring or checking text with calibrated confidence. Because nothing is generated, the answer never needs parsing and cannot hallucinate.

How do I add it to a project?

Add laya-candle = "0.1" and serde_json = "1" to your dependencies. Rust 1.88+ and a C compiler are required to build, because candle-core enables the tokenizers onig feature, which compiles Oniguruma. It is a library only, with no CLI.

Where do the model weights come from?

The crate loads Laya's checkpoints directly from the Hugging Face Hub via candle, and can also load from a local directory. There is no Python involved anywhere in the path.

How does batching work?

All questions are answered in a single batched forward pass, so cost scales with input length, not output length. The primitives are choice (pick one label), score (rate against ordered levels) and noul (a yes/no probability).

Can I gate on confidence?

Yes. Every answer carries a calibrated confidence, and below_confidence lists the answers under a threshold so you can escalate them to a human; presets::guard() is a ready-made prompt guard for existing middleware.

What does a typical call return?

From the README example, a support message yields department Some('billing') with 0.86 confidence, an urgency score of 1.44, and a churn risk of 0.825. Agent is immutable and predict takes &self, so one checkpoint can be shared behind an Arc.

Jev directory

CommunityJev Repos & SDKs

Laya Candle

A pure-Rust port of Laya that answers typed questions — choice, score or yes/no — about any text in one batched forward pass, loading checkpoints from the Hugging Face Hub via candle and returning calibrated labels, scores and probabilities with no Python.

See highlights Open github.com

Stars: 10
Language: Rust
License: APACHE-2.0
Reading time: 2 min
Published by: Community
Author: trystan-sa
Last verified: Oct 2026

trystan-sa/laya-candle

User repository on GitHub

View on GitHub

JEV-like Rust port using Laya, 100% local, no Python.

Forks: 4
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

Tagscommunity catalog-import madewithjev rust local-first calibration

Works forJev for Classification Jev for Scoring Jev for Decision gating

Used inJev for Software & DevTools Jev for Customer Support

Languages: Rust
Capabilities: Open sourceRuns locally

Highlights

Choice, score and noul questions are answered in a single batched forward pass; cost scales with input, not output.
Loads Laya checkpoints directly from the Hugging Face Hub via candle, or from a local directory or JSON schema.
Agent is immutable and predict takes &self, so one checkpoint behind an Arc serves every worker of a service.
below_confidence(0.85) lists answers to escalate; nothing is generated, so answers never need parsing.
Rust 1.88+ with a C compiler required; Apache-2.0, published on crates.io as version 0.1.

Quickstart

rust

[dependencies]
laya-candle = "0.1"
serde_json = "1"

Watch out

Grounded on a truncated README and GitHub metadata captured 2026-10-15; the README's middleware example is cut mid-code, so the Arc pattern is described from the visible portion. The crate is a library only, with no CLI, and the digest carries no accuracy or latency benchmarks.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

The Open Reimplementations: Local Models That Speak System One

Sep 22, 2026

Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.

Jev Repos & SDKs#open-models#research#local

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai