For people who ship faster than they can name folders.
Your hard drive is a graveyard of half-finished ideas. Somewhere in there is a project that was 80% done, one that solved a problem you're about to solve again, and three you have completely forgotten exist.
Workbench scans your drives, identifies every project by framework, runs the ones you trust, screenshots them, and turns a decade of scattered folders into something you can actually browse.
Scan → Identify → Run → Screenshot → Browse → Decide → Act
Finds everything. Point it at ~/Code, ~/Projects, an external drive — whatever.
It walks the filesystem and recognises Next.js, Vite, Tauri, Rails, Go, Rust, Python,
Chrome extensions, Godot, WordPress, and more from their signature files. It knows the
difference between a Tauri app and the Vite app inside it.
Shows you what they look like. For projects you explicitly trust, Workbench starts
them, waits for the server, captures desktop and mobile screenshots, then kills the
process. No more wondering what test-app-4-final-new was.
Tells you what's closest to done. A ship score built from real signals — does it run, does it have a README, auth, payments, deployment config, tests, recent activity — with an effort estimate.
Remembers your whole career. A contribution heatmap built from local git history across every scanned repo, so private and unpushed work counts too. Token usage from your Claude Code and Codex logs. Timelines that merge commits with scan activity.
Browses files properly. Cached Miller columns, folder favorites, instant previews for code, Markdown, images, video and PDFs, git status in the gutter, and build artifacts folded away so a repo is readable at a glance. 007 Mode recursively gathers Markdown and PDF files into a draggable, presentation-ready dossier wall.
Keeps the library flexible. Switch between grid and visual list layouts, choose the column and page count, filter by status or your own project tags, and keep those choices between launches. Media has persistent filters, pagination, fit-to-frame previews and a three-second safety check before removing favorites. Heart images directly from File Finder, paste a project URL to save it instantly, or open every saved link from one list.
Finds the videos buried in a project. The Videos tab recursively gathers YouTube, Vimeo, Wistia, Voomly and Loom links from project files, skips dependencies and build output, and lays them out as playable cards with an open-in-browser fallback.
Manages your agents. Detects which AI coding CLIs you have installed — Claude Code, Codex, Gemini, Cursor, Copilot, OpenCode, Crush, OpenClaw, Hermes, Pi and more — and manages the skills installed for Claude Code and Codex, with search and install from skills.sh built in.
Starts the next one. 127 verified project starters across React, Next.js, shadcn/ui, Vue, Svelte, Astro, Python, Rust, Go, Ruby, PHP, mobile, desktop, extensions and AI.
Runs your agents in-app. Launch Claude Code, Codex or any installed CLI straight into a project from an embedded terminal that floats, moves, resizes and survives navigation. When a project won't start, hand it to an agent — it investigates, reports back a working run command and URL, and Workbench saves them. Frameworks that can't be served headlessly (Chrome extensions, Godot) get manual instructions instead of a dead spinner.
Watches your live services. Optional plugins pull Railway deployments, unresolved Sentry issues and open GitHub pull requests into one dashboard. Pick the projects, repos and authors you care about; tokens stay in your local database. Any Sentry issue can be sent to a coding agent as a fully written prompt with the stack trace attached.
Speaks MCP, both ways. Browse every MCP server configured across your agents in one
place, and expose your own catalog as a server — list_projects, get_project and
library_stats — installable into Claude Code, Codex, Gemini CLI or Cursor with one
click from Settings. Servers you do not need can be hidden from Workbench without
changing another tool's configuration.
Stays current. Workbench checks its public GitHub releases automatically and keeps an Update available button visible when a newer build is ready. The sidebar collapses to an icon rail when you want more room, while keeping the current version visible in the expanded view.
Download the latest .dmg from Releases, drag Workbench to
Applications, and open it.
Requires macOS 10.15 or later. Optional: Chrome for screenshots, git for history, tmux for background AI sessions, Node for JavaScript projects. Workbench checks for these on first run and tells you what each one unlocks.
Everything lives in ~/.workbench/ — a SQLite database, screenshots, and generated
prompts. Nothing is uploaded. There is no account or telemetry. Network access is
limited to update checks plus the services and actions you choose to enable.
Workbench never executes your project code without permission. Scanning is strictly read-only. Running a project requires you to approve the exact command first, and the trust flag is per-project.
git clone https://github.com/WynterJones/Workbench.git
cd Workbench
npm install
npm run tauri devRequires Rust and Node 20+.
npm test # frontend tests
cargo test # in src-tauri/
npm run tauri buildsrc/ React 19 + TypeScript + Tailwind v4 + shadcn/ui
components/ app shell, shared primitives
features/ intro, library, project, files, media, portfolio,
skills, models, plugins, mcp, terminal, timeline, settings
hooks/ data access
lib/ types, api, store, formatting
src-tauri/src/
db.rs models.rs SQLite persistence
scan/ walker, framework detection, git, metadata
run/ process manager, screenshot capture
files/ listing, path guard, starters, context
plugins/ Railway, Sentry, GitHub pull requests
skills.rs agents.rs skill and agent management
pty.rs handoff.rs embedded terminal, agent run-fix sessions
mcp.rs mcp_server.rs MCP config browser, Workbench's own MCP server
score.rs heatmap.rs ship score, contribution graph
Tauri v2 with a Rust core. Dark theme only.
MIT
Made by Wynter.ai