Is this another LLM agent loop, or something genuinely different?

There is no generated text anywhere. You declare the action space in YAML or Python, and the harness compiles it into typed choice, noul, and score questions. The model answers the action, its parameters, and the goal check in one request per step, and every run ends with a structured reason.

Can it type arbitrary text into a field, or only pick declared options?

Only declared options. Free text is rejected outright because a System One model does not generate text. Parameters must come from fixed choices, observation candidates, a boolean flag, or ordered levels. In the browser environment, text comes from named values you supply with --text and the model picks a value by name.

Are those benchmark numbers from real runs, and what do they leave out?

They are live runs: five per scenario against a pinned Jev version through OpenRouter, with all three order-fulfilment goals met on 5 of 5 and mean model latency between 197 and 241 ms. The README explicitly says the results are not claimed for ambiguous state, arithmetic, dates, or long irrelevant context.

What stops it acting on a low-confidence decision?

Every decision carries a probability, and the controller gates on the weakest required one. You set separate thresholds for read, write, destructive, and finish actions in the action space. A run that cannot clear its gate ends as incomplete, failed, or cancelled rather than acting anyway.

How do I run the built-in example?

Clone the repository, pip install -e ., set OPENROUTER_API_KEY or TYPESAFE_API_KEY, then run s1 run --env order:ship_fastest_gift. It drives the order fulfilment environment against the live model and prints each chosen action with its weakest probability and latency, ending with status, wall time and cost.

Which environments can I connect?

Four boundaries: an action space plus a local process you own, an MCP server whose tools are compiled into actions, a browser driven through visible DOM controls, or a Python subclass of Environment implementing observe() and execute(). The same controller handles all four.

How are the confidence gates configured?

An action space marks each action's risk as read, write or destructive, and a gate block sets a probability floor per class, for example read 0.5, write 0.6, destructive 0.8 and finish 0.5. The weakest required probability in a step must clear the selected action's threshold before the environment applies it.

CommunityJev Tools & CLIs

System One Harness

A Python harness for System One models. It runs Jev and other decision models locally, or through HarnessRouter.ai.

See highlights

Open github.com

Stars: 209
Language: Python
License: APACHE-2.0
Published by: Community
Author: HarnessRouter
Last verified: Oct 2026

HarnessRouter/SystemOneHarness

Organization repository on GitHub

View on GitHub

The System One harness for System One models. Run Jev and other System One models locally or directly on HarnessRouter.ai.

Forks: 20
Last pushed: Updated Sep 2026

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

About this project

The harness for System One models. Turn a System One decision model into an agent loop. System One Harness observes an environment, compiles its finite action space into typed questions, gates each decision by confidence, executes the chosen action, and records the complete trace. One model call per step. No generated actions. A probability on every transition. Jev playing a live browser game through System One Harness — one typed decision per step, with no generated control text.

Quoted from the project's README. Read the full README.

Tagscommunity python open-models agents

Works forJev for Routing Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python
Capabilities: Open sourceRuns locally

Highlights

Apache-2.0. 194 stars on 3 October 2026. The README badge reads version 0.3.1.
The description says it runs Jev and other System One models locally or on HarnessRouter.ai.
A conformance report is linked from the README badge.

Watch out

The hosted path is a third-party router, not the TypeSafe API. Check which model a run actually called.

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

jcode

20.4k

Rust coding-agent harness (MIT, ~20k stars) whose crates/jcode-base/src/jev.rs is a typed TypeSafe Jev decision transport, used for memory pruning (memory_jev.rs), browser navigation (browser_jev.rs), and voice routing (voice_intent.rs) with separate provider selectors per purpose.

Jev Tools & CLIs#community#rust#coding-agent

Community

1jehuang/jcodeRust

Strands Decider 2B

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.

Jev Models & Reimplementations#community#open-models#local

Community

TradingAgents

110.5k

Multi-agent trading research with optional Jev screening of StockTwits and Reddit posts for company relevance before sentiment analysis runs.

Jev Tools & CLIs#community#python#trading

Community

TauricResearch/TradingAgentsPython

Awesome Jev Skills

595

A collection of Jev demos, workflows, and agent skills for coding agents: five installable skills and 108 scenarios, installed by pointing Codex, Claude Code, or OpenCode at an agent prompt.

Jev Tools & CLIs#community#python#skills

Community

wuyoscar/jev-skillPython

Harness Router

MIT-licensed Python tool router for coding agents that keeps obvious calls on a fast path, asks Jev (default typesafe/jev-1.13 via OpenRouter) for genuinely ambiguous choices, and can run bounded MCTS when multi-step consequences matter. Ships a CLI, MCP server and hook integrations.

Jev Tools & CLIs#community#routing#mcp

Community

Protocol-Lattice/harness-routerPython

judgekit

CLI and Python library that runs declarative YAML judgment tasks - classify, route, score, verify - natively on Jev Choice/Score/Noul, on any OpenAI-compatible backend, or on local keyword rules, with confidence gates, fallbacks and per-decision cost records.

Jev Tools & CLIs#community#yaml#classification

Community

lexingtonhibiki/judgekitPython

From the Jev guides

Original write-ups on what this entry is for.

Jev on the Gateways: Vercel, Workers AI, and OpenRouter

Oct 12, 2026

The request shape barely changes across surfaces; the surrounding contract does. When to call the API directly and when a gateway, a Worker binding, or a router entry earns its extra hop.

Jev Tools & CLIs#gateway#vercel#cloudflare

Read article

Exporting Decision Models: GGUF, ONNX, and Device Runtimes

Oct 10, 2026

Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.

Jev Tools & CLIs#open-models#local#onnx

Read article

Records at Rest: Documents, Inboxes, and Risk Queues

Sep 28, 2026

Jev labels a record that already exists. Code moves, flags, or holds it. How the document, mail, and transfer tools keep the action out of the model.

Jev Tools & CLIs#classification#email#security

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai