Onevium 是一个以项目为单位的桌面 AI 工作台。它把对话、网页浏览、文件管理、终端和代码审查放在同一界面里,让 AI 完成任务后结果直接显示在旁边,方便继续追问和修改。你可以让 AI 写代码并打开网页预览、分析数据并生成图表、从已有回复分叉出新对话探索别的方案,或用 Git worktree 隔离开发任务。它还内置浏览器,可为不同标签页指定独立身份,并支持从 Chrome 等浏览器导入 Cookie 和密码。通过 Skills 保存可复用的工作流程,用 MCP 和插件接入外部工具,安排定时任务,并把飞书、钉钉、Discord、微信作为对话入口。图像生成需另行接入 OpenAI 或 Gemini 服务。目前提供 macOS 与 Windows x64 安装包,需登录账号或激活许可。
![]()
A project-based AI workspace. Write code, browse the web, create images, and work with files—with the conversation and its results together.
Projects & conversations · Browser · Files & Review · Image creation · Skills & automation
Download stable · Quick install · Documentation · 简体中文

Onevium 1.3.1 is available. Read what's new and what changes when you upgrade. Screenshots were captured during 1.3.0 development using real interface components in an isolated demo, with synthetic projects, conversations, and reports.
macOS · Apple Silicon / Intel — paste this line into Terminal:
curl -fsSL https://raw.githubusercontent.com/Onevium/Onevium/main/install/install.sh -o onevium-install.sh && sh onevium-install.sh
The helper detects your architecture, downloads the official stable installer, checks its SHA-256, and opens the installer. Drag Onevium into Applications to finish. Quit a running copy of Onevium before updating.
Windows x64 — run in PowerShell:
$p = Join-Path $env:TEMP 'onevium-install.ps1'; Invoke-WebRequest 'https://raw.githubusercontent.com/Onevium/Onevium/main/install/install.ps1' -OutFile $p -UseBasicParsing -ErrorAction Stop; & $p
This saves and runs the helper under your existing execution policy, verifies the installer, and opens its setup wizard. The Windows command has not yet been tested on a Windows host; use a direct download if your policy blocks scripts.
| Platform | Current stable installer |
|---|---|
| macOS · Apple Silicon | Download arm64 DMG |
| macOS · Intel | Download x64 DMG |
| Windows · x64 | Download EXE |
The current stable release has no Linux or Windows arm64 installer. Inspect the helpers, choose a version, and read installation details · Compatibility and signing notes
Keep the conversation, web pages, files, and previews for a task in one project. Ask AI to do the work, inspect the result beside it, and continue with the next change.
| What you want to do | How Onevium helps |
|---|---|
| Build a product page | Discuss requirements in the project, edit files, open a web preview, and inspect the changes in Review. |
| Prepare a data report | Work with files, analyze the data, put the summary beside its charts, and ask what to do next. |
| Explore another approach | Branch a new conversation from an existing reply; use a separate Git worktree for development tasks. |
| Create and edit images | Connect an image service, start from a template, prompt, or reference on the Images page, then mark regions and add notes to guide the next edit. |
| Check a server or a database | Save the host or connection in Secrets and hosts, then let AI run the commands you approve, without pasting passwords into the conversation. |
| Continue everyday work | Keep context in projects, reuse processes with Skills, and schedule the next run. |
Keep server, database, API, and cloud credentials in Settings → Secrets and hosts, encrypted with your system keychain. The AI refers to an item by name; Onevium fills in the value on your computer when a command runs, masks known secret values in output, and follows your approval settings. Bring in values from existing config files and CLI accounts without retyping them, and store one-time code seeds so codes are filled in for you.
AI use is off until you turn it on. Approval settings include time-limited grants and optional periods without prompts; ordinary unanswered requests default to allowing one use after two minutes, which you can change to deny. Credential files such as SSH private keys and cloud CLI configs are protected too: the AI has to ask on your computer before reading them.
Add a server or import it from ~/.ssh/config, check its fingerprint on the first connection, and let the AI run commands or read and write files there. Approve once, for a few minutes, or for the whole session, and stop at any time; locking the screen ends every approval.
Open Images in the navigation rail to start from a template or a style, choose the service, aspect ratio, resolution, and number of images, and keep every result in My work. When you want a change, continue the original conversation beside the image.
Image creation needs a separately configured image service.
Curtain reveal now uncovers text word by word, cards ease in as they arrive, and a finished reply stays in place when its steps fold. Switching conversations shows them in one step.
In 1.3.0: more ways to arrange your work
The new navigation rail, sidebar, and tabs bring projects, conversations, and results together. Browser pages, files, terminal, and Review each have a place. When a preview needs more room, dock the conversation in a corner and continue through its floating panel.
Inspect the current turn's changes, branch differences, or the whole workspace. Add comments beside code lines and ask AI to continue. For projects with multiple Git repositories, select the relevant repository or isolate a task in a worktree.
Create a new conversation from an existing reply to explore another approach while keeping the original discussion. Preview the rewind scope, then choose to rewind just the conversation or also restore recorded file changes. File restoration may be unavailable when history is missing or conflicts exist.
Choose an identity for each browser tab to use work and test accounts on the same site. Set the default identity for new pages, or clear one site's data without clearing everything else.
On macOS, import cookies and saved passwords from Chrome, Edge, or Brave, and manage saved passwords in the browser settings. Importing cookies does not guarantee a signed-in session; sites may still require a fresh login or verification.
Connect a supported OpenAI or Google Gemini image service and bring prompts, references, and edit requests into the conversation. Mark regions, add notes, compare before and after, browse previous versions, and save the result you want.
Image creation needs a separately configured image service and its tools enabled. Available models and costs depend on the connected service.
“Analyze this week’s store data, chart the revenue trend, and suggest one experiment for next week.”
Start with a small table. Put the summary, trend, and channel comparison in the same workspace, then discuss what the evidence means.

Sales, orders, channels, and conclusions in this example are synthetic. No live store is shown.
Warm paper, Neutral light, Nord, or Catppuccin: keep the sidebar, conversation, and preview in tune. Here is the same synthetic report in four built-in palettes.

This repository contains product information, demo assets, public installation helpers, and Release installers. It does not contain the application's source code. See Releases for the changes in each published version.
内容由编辑整理,部分信息来自 GitHub 与项目官网,版权归原作者所有。发现错误?报告问题