Yes, a local one: strands-decider serve exposes /v1/systemone, and the Python package loads the checkpoint directly for cuda, mps, or cpu. The card warns the local server has no authentication.

How big is the strands decider?

A 2B-class LoRA adapter plus a readout head on Qwen3.5-2B-Base. The card's recipe ran 1,685 seconds of training on a p5.48xlarge, and documents reproducing it on eight H100s in about an hour or one RTX 3090 in roughly 11 hours.

How accurate is Strands Decider 2B on JevBench?

167 of 231 correct at a 4,096-token window, with Brier 0.348 and ECE 0.050. The card also warns questions are read less reliably than documents and long multi-step documents fall well below the easy tiers.

Can I retrain the decider on my own data?

Yes. The card documents the recipe end to end: about 1 hour 10 minutes on eight H100s or roughly 11 hours on a single RTX 3090, with Qwen3.5-4B's frozen output distribution as the teacher rather than a hosted model.

Is the local server safe to expose?

No. It binds 127.0.0.1 with no authentication, and the card scopes it to local experiments only, so do not put it on a shared or public network.

What hardware can run the decider?

The Python package loads the checkpoint directly for cuda, mps, or cpu, so an Apple GPU or a CPU-only box works as well as an NVIDIA one; strands-decider serve then exposes /v1/systemone.

CommunityJev Models & Reimplementations

Strands Decider 2B

Q: Should I trust its noul and score answers?

Carefully: the card reports noul and score transfer poorly outside the training mix, and calibration is one temperature per primitive fitted on held-out short classification, so measure your own traffic before setting thresholds.

Strands Agents' small decider for agent plumbing: routing, tool selection, triage, and guardrails from a 2B LoRA checkpoint that serves /v1/systemone locally or through a Python package.

See highlights

Open huggingface.co

Published by: Community
Author: Strands Agents
Last verified: Oct 2026

Tagscommunity open-models local agents

Works forJev for Routing Jev for Tool calling Jev for Decision gating

Used inJev for Software & DevTools

Languages: Python
Capabilities: Open sourceSelf-hostedRuns locally

Highlights

JevBench public: 167 of 231 correct at a 4,096-token window with Brier 0.348 and ECE 0.050.
The local server binds 127.0.0.1 with no authentication — the card calls it for local experiments only.
Retraining is documented end to end: about 1 hour 10 minutes on eight H100s, or roughly 11 hours on one RTX 3090.
Its teacher is the frozen output distribution of Qwen3.5-4B, not a hosted model.

Quickstart

bash

pip install strands-decider
strands-decider serve --checkpoint StrandsAgents/strands-decider-2B-hobson-v19 --port 8000

curl -s localhost:8000/v1/systemone -H 'Content-Type: application/json' -d '{
  "state": "The user asks to delete their account.",
  "questions": {"route": {"type": "noul", "instructions": "Does this need human review?"}}
}'

Watch out

The card's own weak spots: questions are read less than documents, long multi-step documents are far below the easy tiers, and noul and score transfer poorly outside the training mix. Calibration is one temperature per primitive fitted on held-out short classification, so measure your own traffic.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

LinkedIn: What multimodal decision models actually change, and what they do not.

LinkedInJevDirectory.org

Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.

jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker News

Hacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.

1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker News

Hacker News: A one-token logprob wrapper that also classifies webcam frames.

allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker News

Related terms

Glossary definitions related to this entry.

Open weights

More Jev resources like this

AutoTrust JEV-27B

An open distillation of Jev 1.13 onto Qwen3.8-27B: a 108.9M-parameter decision block answers typed questions in one prefill pass and adds a POST /v1/decide route beside a full System 2.