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
Umbra 是一款桌面应用,集合日常开发者工具——JSON、Base64、UUID、哈希、JWT、cron 和 OCR——仅公开一个网络例外:更新检查。
AI 翻译
Umbra is a desktop app bundling everyday developer tools — JSON, Base64, UUID, hashing, JWT, cron, and OCR — with one disclosed network exception: the update check.
开发工具TypeScript · RustGitHub @dipaneb在线运营者未确认
试用过了吗? 点一下就好,无需登录。
面向 macOS 的隐私优先开发者工具箱——您的数据永不离开本机。使用 Rust、Tauri 和 Vue 构建。
AI 翻译
A privacy-first developer toolbox for macOS — your data never leaves your machine. Built with Rust, Tauri, and Vue.
取自仓库 README
正在加载完整 README。 以下为已保存的摘录。
A privacy-first developer toolbox for macOS — your data never leaves your machine.
Built with Rust, Tauri, and Vue.
Umbra makes zero network calls except one, explicitly disclosed: the automatic check
for app updates. Installing an update always requires your explicit confirmation
first — nothing installs silently. There is no telemetry anywhere in the app.查看完整 README ↗A privacy-first developer toolbox for macOS — your data never leaves your machine.
Built with Rust, Tauri, and Vue.
Umbra makes zero network calls except one, explicitly disclosed: the automatic check for app updates. Installing an update always requires your explicit confirmation first — nothing installs silently. There is no telemetry anywhere in the app.
The _bmad/, _bmad-output/, and .claude/ directories in this repo are the
BMad Method, a structured
workflow for planning software with an AI agent instead of ad hoc prompting.
It breaks the process into distinct phases — product brief, PRD, architecture
spine, epics, and stories — each with its own reviewable output, rather than
one long unstructured chat.
I ran this project's entire planning phase solo through BMad before writing
any application code: _bmad-output/planning-artifacts/ holds the product
requirements and architecture decisions, and _bmad-output/implementation-artifacts/
holds the resulting story files that drive development one at a time.
.claude/skills/ and _bmad/ are the BMad Method installation itself —
the actual skills and scripts that produced those documents, included so
the process is reproducible, not just its output.
Development (the code in src/, src-tauri/, etc.) follows in later commits.
docs/ holds standalone reference documentation meant to outlive any single story or
PR — for example release/signing setup notes written for future recall, not just
implementation history. It's distinct from _bmad-output/, which records the planning
and story-by-story development process itself.
Anything written there must never include real personal information (full names, local
filesystem paths revealing a username, etc.) — this repo is public. Use placeholders
and generic examples instead; see .claude/hooks/guard-commit-privacy.sh for the
automated guard against this.
Beyond the epics already shipped, a public backlog of candidate features and
improvements is tracked as
GitHub Issues labeled backlog-candidate.
These are candidates, not commitments — the plan is to pick up one small
tool or improvement roughly every week or two from September 2026 through
March 2027.