baiagent-director README

项目的 README,来自 github.com/Balanov/baiagent-director。合并为一个 markdown 文件:README.md。

BAIagent — a company website becomes sourced research and a measurable pilot plan, with the real application result

A free AI advisor that turns a company website, market evidence and local business documents into sourced hypotheses and a measurable pilot plan.

Capabilities · Boundaries · Install · Web audit · Русский

BAIagent first investigates what the company sells, customer feedback, competitors and market signals. It then helps define a hypothesis, actions, owners, timing, budget assumptions, a success metric and a stop condition. The Windows app also supports selected local documents and a local Qwen model without requiring a cloud model API key.

Download BAIagent 1.3.0 for Windows x64 · Release notes · Web audit · Russian documentation

SHA-256: 3ffae27317bb26b6a620ea04990b4b8ec7ddaa0457ef3e8f9b064d16dace52be

BAIagent workflow from entering a company website to a sourced hypothesis and measurable pilot, using real application screens

Real application screens: website input, research result and pilot plan.

BAIagent research result

What it does

BAIagent workflow: website, evidence, hypothesis, measurable pilot and stop condition, with an advisory rather than autonomous boundary

Sources before suggestions. Evidence remains separate from assumptions and recommendations.

Important boundaries

BAIagent is an advisor, not an autonomous employee. It does not independently execute operations in a CRM, publish changes to business systems or guarantee commercial results. A quoted source confirms that text was found; it does not prove that the source's claim is true or that an idea will be profitable.

Fresh market research requires internet access. Saved materials, selected local documents and the local desktop dialog can be used without a network connection. Public website research runs through the web research backend; local document processing and the local Qwen dialog are Windows-app capabilities, not claims about browser-based local inference.

Some websites block automated access, require JavaScript challenges or expose too little public information. In those cases the product should show a warning or an empty result rather than invent competitors, reviews or offerings. Availability of every website, country or market is not guaranteed.

Install on Windows

  1. Download the Windows x64 installer from GitHub Release 1.3.0.
  2. Verify the SHA-256 shown above if you need an integrity check.
  3. Run the installer and follow the setup steps.

The installer and application are not Authenticode-signed. Windows may therefore display an unknown-publisher warning. The package metadata identifies the author as Anton Balanov, but metadata is not a digital signature.

If no compatible local Ollama model is available, the app can download approximately 2.4 GB of model data on first use. Keep sufficient free disk space; 8 GB RAM or more is recommended.

Release 1.3.0 updates BAIagent Director only. Balanov.PRO Enterprise remains at 1.1.1 and is not rebuilt or replaced by this release.

Development

Node.js 24 and npm are required.

npm ci
npm run electron:dev

Validation and web builds:

npm test
npm run build
npm run build:xray
npm run build:advisor
npm run build:enterprise

Desktop packages:

npm run dist:win
npm run dist:linux  # on Linux
npm run dist:mac    # on macOS

The current JavaScript regression suite contains 391 tests in 31 files. Additional PHP suites cover the public research route, international extraction, market filters and sector-specific regressions. Live smoke research has been run for Atlassian in English, Correos in Spanish and Midea in Chinese; these checks are not a claim that every language, market or website has been verified.

The GitHub workflow can build Director for Windows, Linux and macOS. It does not prove manual verification on every platform, macOS notarization or Authenticode signing. The Windows artifact linked above is the verified 1.3.0 build documented in this release.

Web deployment

npm run build:xray

Deploy the contents of dist/ to /xray/ on PHP 8.1+ with cURL, DOM and SimpleXML. The browser calls /xray/api/research.php; no cloud model key is required for that route.

Repository structure

src/             React interface and research results
electron/        Isolated desktop runtime, local files and Qwen
shared/          Research, evidence, recommendations and pilots
public/api/      PHP public research backend
scripts/         Build and verification scripts
tests/           Automated regression tests
android/         Native Android prototype
docs/images/     Documentation images

Security and license

See SECURITY.md, ROADMAP.md and LAUNCH.md.

Source code: AGPL-3.0-or-later. Qwen and third-party dependencies retain their own licenses.

baiagent-director (Balanov/baiagent-director)

正在搜索数据库…

这是什么?BAIagent Director 是一款免费的 AI 商业顾问应用。它可以根据网站资料研究一家公司及其市场,区分事实与假设,标注信息来源,并帮助制定可衡量的试点方案。它不是自主运行的员工。
这是能自动操作我 CRM 的自主智能体,还是只提供建议的顾问?它是顾问,不是自主员工:会输出经过研究的假设和试点方案,但不会在你的 CRM 中实际执行操作。
在 Windows 上怎么把它跑起来?从项目的 GitHub Release 下载 BAIagent Director 1.2.0 x64 安装程序并运行即可;如需企业模式,另有单独的 Balanov.PRO Enterprise 1.1.1 安装程序。
我需要先有云端 API 密钥或安装 Ollama 吗?不需要云端 API 密钥。它可以通过你已安装的本地 Ollama 调用 Qwen,也可以在首次运行时自行下载 Qwen3-4B 模型(约 2.4GB)。建议内存 8GB 以上,并预留足够的磁盘空间。
安装时有什么需要注意的问题吗?Windows 安装程序没有经过 Authenticode 签名,因此 Windows 会提示发布者未知。另外,引用的证据只能证明某段文字存在于来源中,并不能证明该说法属实,也不能证明这个想法有商业价值。
只能在 Windows 上使用吗?主要桌面应用以 Windows 为重点(支持本地文档),但发布流程同时会构建 Director 应用的 Linux 和 macOS 安装包,另外还为审计模式提供单独的网页部署版本。
我的公司数据会留在本地吗?文档和对话记录都在 Windows 本地存储和处理,可离线使用;只有实时市场调研需要联网。
Windows 版应用能在本地读取哪些文档格式?除本地聊天外,它还支持 TXT、Markdown、CSV、JSON、PDF 和 DOCX 文件。
BAIagent、网页审计和 Balanov.PRO Enterprise 有什么区别?它们共用一套代码库,但入口和服务器配置不同:BAIagent 是免费的本地顾问,网页审计通过 PHP 后端运行,Balanov.PRO Enterprise 则是功能扩展的企业审计模式。
它是怎么组织建议的?它按照“观察 → 机制 → 行动 → 验证”的链条构建因果假设,你可以在它起草试点方案之前,选择或添加自己的假设。
有哪些类似的项目?svoi, stemkit, SCM, yovoice, easyread, neo, get-it, TwinDesk
作者
Balanov
大小
6.4 MB
星标
116
时长
21天
提交 · 90 天