Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

sys1

Rust server on candle that serves open decision models like Laya and Clef behind a System One-compatible API with dynamic batching.

Servicebeta
View on GitHub ↗

Overview

sys1 is a self-hosted Rust server for System One open decision models, exposing a TypeSafe-compatible /v1/systemone (or /v1/decide) API so the official Python and JavaScript SDKs can point at it. It is built on tokio, axum and serde with candle for inference, uses dynamic token-based batching, SDPA on CPU, Metal and CUDA, and Flash Attention on recent NVIDIA GPUs, and targets fast inference for ModernBERT and Qwen3.5 architectures. Multimodal decisions from text, images and video work when the model supports them.

Supported models include convaiinnovations/laya (English text, guardrails, email triage), laya-multilingual (100+ languages, about 2.2x faster), laya-typed-decisions, and Cloudflare's clef and clef-flash for text, image and video decisions. A request like routing a refund message returns a choice, an urgency score and an escalate noul from one call.

What the decision model decides

TypeQuestionAnswers
ChoiceWhere should this ticket go?billing · bug · account
ScoreHow urgent is this message?routine, no rush · today · urgent · critical, about to churn
Yes / noEscalate to a human immediately?—

How it works

  1. cargo install sys1 with the right backend feature
  2. Run sys1 --model-id with a supported model
  3. POST typed questions to /v1/systemone
  4. Read typed answers; use official SDKs with a local base URL

Strengths

  • Fast Rust inference with batching
  • CPU, Metal and CUDA builds
  • Works with official TypeSafe SDKs

Caveats

  • Requires a specific Rust version
  • Limited set of supported models so far

Getting started

cargo install sys1 --features cpu, then sys1 --model-id convaiinnovations/laya --dtype auto.

At a glance

Type
Service
Status
beta
Language
Rust
Runtime
HTTP server (candle)
Has code
yes

Use cases and integrations

How curated lists describe it

  • Rust server on candle for self-hosted open decision models such as Laya, with a System One-compatible API, dynamic token-based batching, and CPU, Metal, and CUDA builds

    AnotiaWang/awesome-decision-models

  • Blazing fast, self-hosted structured decisions for open models with a TypeSafe…

    wh000wh000/awesome-jev-live

  • System One compatible API for open decision models, e.g. Laya, written in Rust.

    hellogumbo/awesome-jev

  • System One compatible API in Rust for open decision models such as Laya.

    MrJev/awesome-jev

Frequently asked questions

Which models does sys1 serve?

Laya variants and Cloudflare's Clef models, with more planned.

Can I use the TypeSafe SDK with it?

Yes, set the base URL to the local server.

Does it support images and video?

Yes, when the model supports them.

Related projects

  • decidealot

    Docker service exposing local Laya, Von, a CLM head or hosted Jev through one TypeSafe-compatible HTTP and MCP endpoint.

    MCP serverbetaPython
  • edgejev

    Converts open Jev reproductions (Laya, Kev, NanoJev, PlayJev) to INT8 ONNX for offline CPU decisions at about 15.6 ms per question.

    ServicebetaPython
  • go-system-one

    Native Go runtime serving Gemma 4 12B as a Jev-like decision model with hand-tuned PTX kernels and no CGo or CUDA toolkit.

    ProjectbetaGo
  • gojev

    Go harness for System One models: hosted Jev, open Kev in-process via llama.cpp, and Cloudflare Clef, with typed helpers and a CLI.

    ServicebetaGo
  • hearim

    Go gateway exposing the Jev /v1/systemone contract over ordinary LLM backends by reading single-token label logprobs.

    ServicebetaGo
  • intern-decision-mlx

    Runs Intern-Decision-0.8B on Apple Silicon behind a local /v1/systemone endpoint that accepts screenshots for agent decisions.

    ServicebetaPython