# Jev Harness

Apixly’s independent MIT Python runtime for bounded Jev agent loops. Not an official TypeSafe AI product.

Distribution: apixly-jev-harness. Python namespace: apixly_jev_harness. CLI: jev-harness / apixly-jev-harness. Version: 0.1.17.

Install: pip install 'git+https://github.com/apixly-ai/jev-harness.git@v0.1.17'. The PyPI name jev-harness belongs to an unrelated project.

Programs provide registered tools and parameter candidates, execute effects and independently verify completion. Jev selects supplied references. Inference is billable; unknown usage remains unknown. No general success or savings guarantee.

Historical browser acceptance uses a local synthetic fixture. Homepage media is a zero-inference screenshot replay. Native desktop completion remains unverified.

## English

- [Homepage](https://apixly-ai.github.io/jev-harness/): Jev Harness identity, visual cases and installation.

- [Python agent task contract](https://apixly-ai.github.io/jev-harness/agent-contract.html): Install Jev Harness, define typed tools and parameter candidates, and run bounded Python tasks with explicit context and independent verification.

- [Context management for Jev agent loops](https://apixly-ai.github.io/jev-harness/context.html): How Jev Harness admits goals, facts, dialogue and tool results, preserves evidence references, and blocks repeated actions against unchanged state.

- [Browser and desktop task adapters](https://apixly-ai.github.io/jev-harness/adapters.html): Connect bounded Jev Harness tasks to headless Camofox, macOS Accessibility or Windows UIA. Read permission, freshness and acceptance limits.

- [Jev Harness architecture and design references](https://apixly-ai.github.io/jev-harness/architecture.html): How the primary AI, deterministic program, typed Jev decisions and independent verifier cooperate in Apixly’s Python task runtime.

- [Acceptance evidence and benchmark limits](https://apixly-ai.github.io/jev-harness/evidence.html): Inspect historical live synthetic browser acceptance, offline A/B reports, measured usage, negative results and unverified desktop completion.

- [Jev Harness FAQ: installation, tools and verification](https://apixly-ai.github.io/jev-harness/faq.html): Answers about Apixly’s Jev Harness identity, exact Git installation, Jev Filter integration, typed choices, billing, browser evidence and execution limits.

- [Jev Harness use cases: open research and investigation](https://apixly-ai.github.io/jev-harness/use-cases.html): Design open research, repository diagnosis and controlled API investigation with AI-authored tools, dynamic Jev choices and independently checked artifacts. Acceptance status stays explicit.

- [Open goals, dynamic tools and Jev delegation](https://apixly-ai.github.io/jev-harness/open-goals.html): Give an AI an open goal, let it write task tools, and let Jev discover the route through dynamic tool and parameter choices. Inspect independent evidence and pending acceptance plans.

## 简体中文

- [中文主页](https://apixly-ai.github.io/jev-harness/zh/): Jev Harness 项目介绍、演示案例和安装入口。

- [Python 智能体任务协议](https://apixly-ai.github.io/jev-harness/zh/agent-contract.html): 安装 Jev Harness，定义类型化工具与参数候选，使用明确上下文和独立结果验证运行有边界的 Python 任务。

- [Jev 智能体循环的上下文管理](https://apixly-ai.github.io/jev-harness/zh/context.html): 了解 Jev Harness 如何准入目标、事实、对话与工具结果，保留证据引用，并阻断相同状态下的重复动作。

- [浏览器与桌面任务适配器](https://apixly-ai.github.io/jev-harness/zh/adapters.html): 将有边界的 Jev Harness 任务连接到无头 Camofox、macOS Accessibility 或 Windows UIA，了解授权、新鲜度和验收范围。

- [Jev Harness 架构与设计参考](https://apixly-ai.github.io/jev-harness/zh/architecture.html): 了解主 AI、确定性程序、类型化 Jev 决策和独立验证器如何在 Apixly 的 Python 任务运行时中协作。

- [验收证据与评测范围](https://apixly-ai.github.io/jev-harness/zh/evidence.html): 复查历史真实合成浏览器验收、离线 A/B 报告、实测用量、负面结果，以及尚未验证的原生桌面完成状态。

- [Jev Harness 常见问题：安装、工具与验证](https://apixly-ai.github.io/jev-harness/zh/faq.html): 了解 Apixly Jev Harness 的项目身份、准确 Git 安装方式、Jev Filter 集成、类型化候选、计费、浏览器证据和执行边界。

- [Jev Harness 使用案例：开放调研与探索式诊断](https://apixly-ai.github.io/jev-harness/zh/use-cases.html): 设计开放调研、仓库诊断与受控 API 排障，让 AI 编写工具、Jev 动态选择并独立核对产物，明确保留验收状态。

- [开放目标、动态工具与 Jev 托管](https://apixly-ai.github.io/jev-harness/zh/open-goals.html): 给 AI 一个开放目标，让它编写任务工具，再由 Jev 通过动态工具与参数选择探索未知路径。复查独立证据与待完成的验收计划。

## Machine-readable evidence

- [Task spec](https://apixly-ai.github.io/jev-harness/spec.json)

- [Historical acceptance](https://apixly-ai.github.io/jev-harness/acceptance.json)

- [Media provenance](https://apixly-ai.github.io/jev-harness/media/showcase.json)

- [Repository](https://github.com/apixly-ai/jev-harness)