neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
翻译准备中
尚未检测到更新。GitHub 发布版本后会显示在这里。
个人 GitHub 账号下公开仓库的所有者可以登录并验证所有权,自行管理介绍,无需管理员批准。
登录并生成验证文件,在 30 分钟内将其提交到此仓库的默认分支。其他分支或分叉仓库中的文件无效。然后在下方检查文件,即可无需管理员批准直接管理项目。请保留文件以维持管理权限。
如需登录,将前往登录页面。登录后请回到此处,再次点击所选的验证按钮。
在项目文件夹中运行 /nomorevibe verify,即可验证所有权并自行管理介绍。
一分钟完成验证MinusX 是一个代理式商业智能平台。它是用于数据的 Claude Code。
AI 翻译
MinusX is an Agentic Business Intelligence platform. It's Claude Code for data.
取自仓库 README
正在加载完整 README。 以下为已保存的摘录。
MinusX is an open source agentic business intelligence platform for founders. Your data stack, staffed by agents.
Website
·
Docs
·
MinusX OSS installation cmd: curl -fsSL https://minusx.ai/install.sh | bash查看完整 README ↗MinusX is an open source agentic business intelligence platform for founders. Your data stack, staffed by agents.
MinusX OSS installation cmd: curl -fsSL https://minusx.ai/install.sh | bash
Note: This is on top of obvious BI features like connecting to databases/warehouses, questions, dashboards, reports, etc.
curl -fsSL https://minusx.ai/install.sh | bash
Requires Docker. Run it in a terminal: while the image downloads, the script walks you through setup — workspace, AI model (with a live connection test), and optionally your first database — so you just log in when it finishes. Run non-interactively and it skips the questions and starts the app, and you set the workspace up in the browser.
MinusX is a single Next.js application — there is no separate backend service.
cd frontend
npm install
cp .env.example .env # then set NEXTAUTH_SECRET=$(openssl rand -base64 32)
npm run dev # http://localhost:3000
The LLM provider is configured in the app after first launch (setup wizard "AI Models" step, or Settings → Models) — there are no LLM environment variables.
Architecture, module documentation and development guidelines live in CLAUDE.md.
We've spent years watching companies struggle with the same problem: they buy expensive "self-serve" BI tools, set it up over months, hire analysts, and build dashboards. And yet, most people in the company can't answer basic questions about their own data without pain and tears.
When LLMs got better, everyone (including us) bolted chatbots / text-to-SQL onto existing tools. Unfortunately, it doesn't work. These tools were designed around menus, isolated query editors and tons of scaffolding. The AI is an afterthought (a parlor trick almost) that breaks the moment you need it to do anything real.
We built a state-of-the-art data agent that learns your business (ranked #1 on DataAgentBench, a realistic data-tasks benchmark from UC Berkeley's EPIC lab). And then built a BI platform around it. Learning from the magic of Claude Code, MinusX exposes the entire BI (questions, dashboards, reports etc.) as a file system the agent can read and write. The agent is omni-present, working exactly how you work.
Anyone who has spent any time working with LLMs in data knows that the hard part is not SQL, it's context. The model doesn't know that revenue is ARR_operational in your company, or that the orders table has a quirk where cancelled orders still show up, but only till 2025(!!!). MinusX Knowledge Base gives you tools to teach the agent what it needs to know, and learns from your continued usage. A 200 line generated SQL that you cannot understand, trust and reason about is as useless as not having the answer.
dbt is amazing. Semantic models are great. Still, less than ~10% of fast growing companies have all their data modeled. This tech-debt only grows as you scale. MinusX is designed to work with or without dbt. Write messy SQL and have the agent clean it up!
MinusX is Open Source Software and licensed under the AGPL-3.0 license. See LICENSE file for details.