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

又一份"本地 vs 托管"选型实录：astgl.com（James Cruce，09-22）—— 421M 开源模型在其 frozen replay 上胜过自制路由；他最终不选 Jev 是因为约束（数据不出本机、可控回退），并主动限定*"whether Laya beats J…

James Cruce's write-up on choosing local LAYA over hosted Jev: the 421M open model beat a home-made router on a frozen replay.

Articlediscussion
Open source ↗

This summary is based on how curated lists describe the project; its README was not available to review.

Overview

This astgl.com post by James Cruce, dated 22 September, records a local-versus-hosted selection. Based on the Chinese list title, the 421M open model (LAYA) beat the author's own router on a frozen replay of his traffic. He ultimately chose not to use Jev because of constraints: data must not leave the machine and fallback behavior must be controllable. He also explicitly limits his claim, not asserting whether LAYA beats Jev in general.

It is a useful example of constraints, not only accuracy, driving the choice between hosted and local decision models.

What the decision model decides

TypeQuestionAnswers
ChoiceRouting decisions replayed against LAYA and a home-made router(inferred)—

How it works

  1. Record real routing decisions as a frozen replay.
  2. Run LAYA and the existing router on it.
  3. Compare and weigh deployment constraints.
  4. Choose local LAYA for data locality and controllable fallback.

Strengths

  • Grounded in the author's real traffic.
  • Explicit about the scope of the comparison.

Caveats

  • Single author's workload.
  • Does not compare LAYA and Jev head to head.

At a glance

Type
Article
Status
discussion
Has code
no
Has tests
no

Use cases and integrations

How curated lists describe it

Frequently asked questions

Why not Jev?

Because of constraints: data had to stay local and fallback had to be controllable.

Did LAYA beat Jev?

The author explicitly does not claim that; LAYA beat his own router.

Where is it?

On astgl.com.

Related projects

  • 第三方独立实测：Classmethod / DevelopersIO

    Classmethod DevelopersIO independent test of Jev for LLM model routing, reporting 40 successful calls and 10 of 10 correct classifications.

    Articleresearch
  • instinct-ai

    Open-source dual-brain runtime that answers typed decisions with a fast local engine or the Jev API before falling back to LLMs.

    ServicebetaPython, TypeScript
  • jev-laya-dsh

    Local Laya decision layer for DeepSeek Harness: routes prompts, gates risky tool calls and swaps models, all offline at no token cost.

    Projectbeta
  • jev-switch

    Unofficial Rust gateway routing /v1/systemone requests to TypeSafe, Vercel, OpenRouter or local Laya via an editable DAG.

    ProxybetaRust
  • Keel

    Local-first macOS coding workspace that runs your agents and lets a local Laya or hosted Jev selector pick a route the host validates.

    ServicebetaRust
  • local-jev-api

    Docker recipes and ARM64 images for serving ten native decision checkpoints, such as Cloudflare Clef, behind /v1/systemone on DGX Spark.

    ServicebetaPython