A local-first AI workspace that keeps your work private, organized, and ready to run.
English | 简体中文
Most AI tools force you to choose between capability and privacy — and leave you rebuilding context from scratch every time. Cowork-Z takes a different approach.
Each project lives in its own workspace: one folder, all your files, sessions, and agent history in one place. Your data never leaves your machine. And a built-in Skills Catalog and Starter Packs mean you're productive in minutes, not hours.
Whether you're a developer protecting proprietary code, a researcher with sensitive data, or a knowledge worker who wants AI on your local files — if privacy matters, this is for you.
Schedule recurring AI tasks that run unattended — and only ping you when there's something worth your attention.
- Set it on a cron — Daily, weekdays, weekly, or any custom schedule. A structured picker lets you configure frequency, day, and time with a live cron preview; switch to Custom mode for direct cron input.
- Runs while you sleep — Each automation goes through the full task pipeline, hands-free. The Rust-based scheduler fires on schedule, dispatches work to the sidecar, and queues runs when the agent is busy.
- Smart triage — Findings surface in a dedicated sidebar tab with unread badges; quiet runs with no findings are auto-archived. Click any run to jump straight to the full agent output.
- Bring your own model — Pin a specific provider and model per automation. Works with all 12+ supported providers.
- Workspace-scoped — Each automation is bound to a workspace. Switch workspaces and the automations list and triage view update automatically.
- Run Now — Trigger any automation on demand from the action menu, bypassing the schedule.
Type one prompt, run 3 models in parallel, and compare results side-by-side. Arena helps you find the best model for a given task without switching back and forth.
- 3 independent agents — each column runs its own OpenCode session with a different model, sharing the same prompt and workspace
- Any model mix — pick from all 12+ supported providers per column (e.g., Claude + GPT + Gemini in one arena)
- Rich output — full Markdown rendering, syntax-highlighted code blocks, tool call cards, clickable links, and streaming in every column
- Tabbed layout — switch between columns with model-name tabs and live status badges
- Follow-up conversations — send additional messages and all 3 agents continue from where they left off
- Sidebar integration — arena sessions appear alongside regular tasks with an expandable disclosure triangle showing each model's individual chat
Your files never leave your machine. The agent reads and writes locally — nothing is uploaded to cloud servers. You control exactly what it can access:
- Folder-level access controls — grant read or read-write per directory
- Runtime permission prompts — if the agent needs a folder you haven't approved, it asks first
- Per-session tracking — permissions are scoped and persisted per task
The agent asks before accessing folders outside your approved list
Connect to 12+ AI providers and switch between them at any time:
Direct API — Anthropic, OpenAI, Google Gemini, xAI, DeepSeek, Z.AI
Cloud Platforms — GitHub Copilot, AWS Bedrock, Azure AI Foundry
Proxy Services — OpenRouter, LiteLLM
Local Models — Ollama
Credentials are stored in the OS Keychain (macOS Keychain, Windows Credential Manager, Linux Secret Service) — never in plain text files.
Connect to any provider — credentials stored in the OS Keychain
Work is organized around workspaces — one folder, one focus. Each workspace holds its own files, chat sessions, permissions, and agent history in one place. Switch workspaces to switch projects; nothing bleeds between them.
- File tree browser — lazy-loaded sidebar tree with real-time filesystem watching and search
- File preview panel — preview code (syntax-highlighted), Markdown, images, video, PDF, and HTML without leaving the app; supports fullscreen mode
- "Add to Chat" — insert any file as an
@pathreference into the chat input directly from the preview panel
One workspace per project — files, sessions, and history in one place
- Inline file previews — file paths render as clickable links with thumbnails for images and video
- Image gallery — click any thumbnail to open an in-app preview with "Show in Finder"
- URL previews — links in agent responses open in your default browser
- Drag-and-drop — drop files or folders from Finder or the file tree into the chat to reference them
- Multi-line input — compose detailed prompts with
Shift+Enter - Todo tracking — see the agent's task progress in a sidebar panel with a progress bar
- Artefacts panel — all files the agent creates or modifies are tracked in the sidebar
File paths render as clickable links with inline media previews
Hit the ground running with 6 guided workspace packs covering writing, research, security audits, legal review, and more. Each pack bundles template files, prompts, and step-by-step guidance:
- Browse packs on the Home screen and click Install
- Choose a destination folder — the app creates the workspace automatically
- The agent opens
START_HERE.mdand walks you through the task
No configuration, no setup from scratch.
Browse and install guided workspace packs from the Home screen
A built-in catalog of reusable AI skill templates, installable with one click:
- Category tabs — browse by domain (Marketing, Sales, Enterprise, and more)
- Search — filter by name, description, or category in real-time
- One-click install — skills are copied to
~/.config/opencode/skills/and auto-discovered by the agent - Update detection — SHA256 checksums flag outdated skills with a Re-install prompt
One-click install of reusable AI skills — no manual file management
A dedicated window for managing skills from Git repositories:
- Register Git repos — add any public or private Git repository as a skill source
- Auto-discovery — skills are automatically discovered from cloned repos by scanning for
SKILL.mdfiles - Browse & install — search, filter, and install skills from registered repos into any global skills directory
- Sync & update — repos sync on app launch; outdated skills show an Update prompt
- Multiple targets — install to
~/.config/opencode/skills/,~/.claude/skills/, or~/.agents/skills/
Connect external tools and data sources via the Model Context Protocol. Supports both local (command) and remote (URL) servers, configurable from the Settings panel.
- Keyboard shortcuts —
Cmd+Nnew task,Cmd+,settings,Escapecancel - Multiple themes — light and dark modes with runtime switching
- Auto-updates — checks for updates on launch with signed, verified bundles
- Cross-platform — macOS (Apple Silicon & Intel) today; Windows and Linux builds available
Switch between light and dark themes at any time
Go to the latest release and download the .dmg for your architecture:
| Chip | File |
|---|---|
| Apple Silicon (M1+) | cowork-z_*_aarch64.dmg |
| Intel | cowork-z_*_x64.dmg |
Windows support (dev) — Cowork-Z now runs on Windows. Full feature coverage hasn't been completed yet, so please expect rough edges and report issues you encounter.
Linux builds (x64 and ARM64) are produced by CI. Check the releases page for
.deband.AppImagefiles.
Note
Early access — Cowork-Z is under active development. Some features may be rough around the edges. Feedback and bug reports are very welcome.
Cowork-Z requires OpenCode as its AI engine. Install it globally:
npm install -g opencode-aiIf the app can't find
opencodeon launch, it will show a dialog with installation instructions. Make sure theopencodebinary is on your shell'sPATH.
- Open Cowork-Z
- Press
Cmd + ,(or click the gear icon) to open Settings - Pick an AI provider (e.g. Anthropic, OpenAI, Google Gemini, Ollama ...)
- Enter your API key — credentials are stored securely in the OS Keychain, never in plain text
Type a prompt in the launcher (or press Cmd + N) and hit Enter. The agent will plan, execute, and report back — all running locally on your machine.
Try these:
"Summarise all PDFs in this folder into a single report""Review this codebase for security issues and write findings to AUDIT.md""Organise the photos in ~/Downloads by date into year/month folders"
- Automations — Switch to the Automations tab on the Home screen. Create a recurring task with a schedule and model, and let it run on its own. Findings appear in the sidebar triage tab.
- Starter Packs — Browse the Home screen, pick a guided workspace pack, and click Install. The app creates a workspace and the agent walks you through
START_HERE.md. - Skills Catalog — Click Install on any skill from the Home screen to add it to your OpenCode skills directory. No restart needed.
- Skills Manager — Open the Skills Manager from the sidebar to register Git repositories as skill sources, browse discovered skills, and install or update them across multiple skills directories.
- MCP Servers — Open Settings > MCP Servers to connect external tools and data sources (databases, APIs, file systems) via the Model Context Protocol.
| Layer | Technology |
|---|---|
| Desktop framework | Tauri 2.x (Rust + Web) |
| Frontend | React 19, TypeScript, Tailwind CSS, Radix UI / shadcn/ui, Zustand |
| Build | Vite, Cargo, pnpm |
| Database | SQLite (rusqlite) |
| Secure storage | OS Keychain (keyring crate) |
| AI Engine | OpenCode via Node.js sidecar |
We welcome contributions! See CONTRIBUTING.md for development setup, coding guidelines, and how to submit pull requests.
git clone https://github.com/kevinlin/cowork-z.git
cd cowork-z
pnpm install
cd src-tauri/sidecar-opencode && pnpm install && cd ../..
pnpm tauri dev- Node.js v20+, pnpm v9+
- Tauri v2 toolchain (see https://v2.tauri.app/start/prerequisites/)
- Rust toolchain (for the Tauri backend)
- OpenCode (
npm install -g opencode-ai)
Tauri on Windows requires the MSVC toolchain. If you see missing dlltool.exe errors or your rustc -vV shows host: x86_64-pc-windows-gnu, follow these steps:
-
Install Microsoft C++ Build Tools
- Download Visual Studio Build Tools
- In the installer, select the Desktop development with C++ workload
- Verify the MSVC v143 toolset and Windows 10/11 SDK are included
- Open a new terminal after install and confirm with
where cl
-
Install rustup (if
rustup/cargoaren't found)- Download and run rustup-init.exe — choose the default install
- Ensure
%USERPROFILE%\.cargo\binis on yourPATH - Verify:
rustup --version && cargo --version
-
Switch to the MSVC toolchain
rustup toolchain install stable-x86_64-pc-windows-msvc rustup default stable-x86_64-pc-windows-msvc
Verify with
rustc -vV— you should seehost: x86_64-pc-windows-msvc. -
Remove any per-repo GNU override (common gotcha)
rustup override list # If your repo is listed with ...-gnu: cd path\to\cowork-z rustup override unset
-
Clean and rebuild
cd src-tauri && cargo clean && cd .. pnpm tauri dev
Note: Windows 11 ships with WebView2 pre-installed. If the build complains about WebView2, install the runtime and retry.
See the requirements document for the full feature spec. Outstanding items:
- Windows testing and polish
- Database encryption at rest (Req 5.2.2)
Track progress on the issues page or check the changelog for recent releases.
- Cowork-Z's UI is based on Accomplish. Thanks to the Accomplish team for their work on the original implementation.
- The Workspace, Starter Packs and Skills Catalog features were inspired by and heavily referenced from Tandem. Thanks to the Tandem team for pioneering this pattern.
MIT — Copyright (c) 2025-present Kevin Lin and contributors