Installation

Install ROW Recall

Signed binaries for Windows, macOS and Linux, the VS Code extension, and the MCP server for AI agents.

Install the desktop app

ROW Recall is distributed as pre-built, signed binaries. The most reliable path is to download the signed installer for your platform from the website or from GitHub Releases.

PlatformBinary / package
WindowsReCall_0.8.0_x64-setup.exe (signed, per-user NSIS installer)
macOSReCall_0.8.0_x64.dmg
LinuxReCall_0.8.0_amd64.AppImage (plus .deb / Flatpak)

Package managers

The documented package-manager install commands are: Download from the website remains the primary, always-up-to-date path.

PlatformOne-liner
Windows · wingetwinget install RowAI.ReCall
Windows · Scoopscoop install recall
Windows · Chocolateychoco install recall
macOS · Homebrewbrew install --cask rowai/tap/recall
Linux · dpkgsudo dpkg -i recall_*.deb
Linux · AppImagedownload ReCall_0.8.0_amd64.AppImage, chmod +x, run

Availability note (accurate as of v0.8.0): the winget manifests validate locally but have not yet been submitted to microsoft/winget-pkgs (pending installer code-signing), so winget search RowAI may return nothing today. If a package-manager install is unavailable, use the signed binary from Download or GitHub Releases — the packages and the GitHub binaries are the same Release artifacts.

First run

  1. Launch the app. A first-run signup card asks for your email (name and country are optional) with privacy consent.
  2. Verify your email with a 6-digit OTP when prompted (if you already verified on the website this is skipped).
  3. Add a workspace with + Add Tracked Folder and pick your project.
  4. Verify your email to receive the 30-day Pro welcome coupon, or enter a Pro key later in Settings.

Install the VS Code extension

The ReCall extension (rowai.recall, v0.8.0) is a standalone extension — it does not require the desktop app or Git. It stores everything in VS Code’s own protected storage using an embedded WASM SQLite engine, so it works offline and locally on Windows, macOS and Linux.

# From the VS Code / Cursor / Windsurf Marketplace # Search for "ReCall" (publisher: rowai) and install, or side-load the .vsix: code --install-extension recall-0.8.0.vsix

Key extension commands (Command Palette, Ctrl+Shift+P / Cmd+Shift+P):

CommandPurpose
ReCall: Create CheckpointSnapshot the workspace with an optional label
ReCall: Rollback to Event...Restore the workspace to an event id
ReCall: Restore This FileRestore the current file to a past event
ReCall: Show Diff for EventShow the unified diff of an event
ReCall: Undo Last OperationUndo the last restore operation
ReCall: Redo Last OperationRe-apply the last undone operation
ReCall: Changed Files Since CheckpointList files changed since the last checkpoint
ReCall: Set Workspace FolderSet which folder the extension tracks

Extension settings are scoped under recall.*: recall.maxDiffFileSizeKB (default 1024), recall.debounceMs (default 200), and recall.maxEventsShown (default 100).

Install the MCP server for AI agents

The @rowai/mcp-server package exposes the ReCall engine as 17 standard MCP tools that any MCP-capable host (Claude Code, Codex, Cursor, Windsurf, VS Code and more) can call autonomously. It is a thin bridge to the local WebSocket daemon.

npm install -g @rowai/mcp-server

Then register it in your MCP host configuration:

{ "mcpServers": { "recall": { "command": "npx", "args": ["@rowai/mcp-server"] } } }

The server requires the ReCall daemon to be running and watching a workspace — launch the desktop app or run recall watch. It connects to ws://127.0.0.1:9876 (override with the RECALL_WS_URL environment variable) and reconnects automatically with exponential backoff. See Integrations for per-tool and per-platform guides.

System requirements

ItemRequirement
OSWindows, macOS, or a mainstream Linux distribution
Runtime (MCP server / agent tools)Node.js 18+ (Node 22+ recommended for hooks)
MemoryThe desktop daemon uses roughly ~24 MB RAM, ~0% CPU when idle
StorageThe timeline is delta-compressed; a typical project is a few to tens of MB