Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

Harnessix Code

本地运行、模型自选、每一步都受控的 Coding Agent。

Harnessix Code 在你的终端里读代码、改代码、跑测试。改动要你批准,命令默认进沙箱,崩溃后能恢复,全过程可审计。模型由你选择,可以是云服务,也可以是你自己部署的模型。

当前是预览版,不是正式商用版本,不提供任何担保。 请在可以通过 Git 回退的仓库里使用。

快速开始

需要 macOS 或 Linux、Python 3.12 以上,以及 uv。

# 1. 安装(国内可加 --index-url https://mirrors.huaweicloud.com/repository/pypi/simple)
uv tool install harnessix

# 2. 配置模型
export MODEL_API_KEY=你的Key
harnessix code configure --provider-kind openai_chat \
  --base-url https://api.deepseek.com/v1 --model deepseek-v4-pro --non-interactive

# 3. 检查并启动
harnessix code doctor /path/to/repo --probe-model
harnessix code /path/to/repo
  • 安装、配置、使用和全部功能的完整说明,见用户手册。
  • 使用百炼等其他服务,或简要的上手说明,见 Beta 快速上手。
  • 使用自己部署的模型(vLLM、Ollama,同机或局域网),见接入自托管模型。
  • 无人值守运行:harnessix exec --trust-workspace --prompt "..." --approval-mode auto-edit。

它能做什么

能力 说明
代码读写 读取、搜索、多文件事务修改,修改前显示完整差异
运行命令 默认在 macOS Seatbelt 或 Linux bubblewrap 沙箱内自动执行,需要联网时逐条请你批准
审批 逐次批准;也可以在本会话自动接受修改,或允许完全相同的操作
恢复 进程崩溃或断线后可继续,同一副作用不会被盲目重放
上下文 按模型窗口自动压缩,读取仓库中的 AGENTS.md
Git 查看状态和差异;你要求时,经批准后提交
Skills 按需加载个人或仓库里的任务说明,见使用 Skills
MCP 在本机启动你逐个信任的 MCP Server,每次工具调用都需批准,见使用 MCP
模型 OpenAI 兼容接口与 Anthropic 接口;已预设 DeepSeek 和百炼,支持自托管模型
集成 终端界面、无头模式(JSON 输出)、stdio App Server、Python SDK

还没有: Hooks,IDE 插件,Web 搜索,子代理,图片输入,自动更新。

安全边界

  • 沙箱内的命令不能联网,不能写 .git 和工作区以外的位置,也读不到 ~/.ssh 等凭据目录。
  • 沙箱外的命令每条都需要你批准。
  • 首次打开目录时需要确认信任。
  • API Key 只从环境变量读取,并在输出和记录中按机密保护。

沙箱内命令能读取拒读清单以外的文件,这些内容可能发送给模型服务商。完整说明和已知限制见 SECURITY.md。

数据去向: 代码和对话只发送给你配置的模型服务商。Harnessix 没有遥测,不向项目方发送任何数据。详见隐私说明。

评测结果

使用同一个本地模型(Qwen3.8-Flash-Next),在 SWE-bench Verified 的 49 题留出集上单次运行,官方判分:

系统 解决题数
Harnessix(当前) 27
mini-swe-agent 29(运行时能访问公网,其中至少 4 题直接抄了上游修复,扣除后约 25)

这个结果目前只能说明两者接近,mini-swe-agent 还需要在断网条件下重跑。此外,这 49 题在调优过程中都被用于诊断,结果偏乐观;评测配置使用了 SWE-bench 任务模板;单次运行的波动也不小。SWE-bench 也不代表你的真实项目。

文档

参与和许可

社区版按照 AGPL-3.0-only 发布。闭源使用见商业许可说明,品牌规则见 TRADEMARKS.md,第三方依赖见 THIRD_PARTY_NOTICES.md。

贡献代码前请阅读贡献指南。发现安全问题请按 SECURITY.md 私下报告,不要公开提 Issue。

Metadata

Release files for harnessix 0.10.0b7

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Built distribution (wheel)

Table of built distributions (wheels) for harnessix 0.10.0b7
File Interpreter ABI Platform Actions
harnessix-0.10.0b7-py3-none-any.whl Python 3 none any Details for harnessix-0.10.0b7-py3-none-any.whl

Release files / harnessix-0.10.0b7-py3-none-any.whl

Download URL harnessix-0.10.0b7-py3-none-any.whl
Size 1.7 MB
Tags Python 3
SHA-256 checksum
How to use checksums
84eac2aca638bdddb7ba65b0e3124d48440cd409de08d4a658580d053aac564c
BLAKE2b-256 checksum
How to use checksums
6102914cf0988979f66285ab8962a2701bc5386e9f318b6a8149d95ce9eb070f
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 11, 2026.

Transparency log

Security for harnessix 0.10.0b7

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

1 file in this release has been uploaded with provenance metadata.

Attestation for harnessix-0.10.0b7-py3-none-any.whl

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 11, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page