Sinclair

The terminal your agent team lives in.

Run a team of coding agents in a window of panes — each on a shared bus, each reporting whether it's working, blocked, or done — on top of a terminal that stands on its own. GPU-rendered, standards-complete VT, tabs and splits, live-reloading config, 22 themes.

macOS and Linux · Rust · Apache-2.0

brew install --cask wess/packages/sinclair

Linux install · Documentation · Tutorials · Trailer

---

The agent workspace

Every other terminal treats a coding agent as just another process that prints things. Sinclair treats it as something you're supervising.

• Status per pane. Every pane self-reports a semantic state — working, blocked, done, idle — as a colored dot on its tab, rolled up in the Activity panel. sinclair agent-hooks install wires Claude Code's lifecycle to it, so you can see which of six agents is stuck without clicking through six panes.
• Whole teams, one window. Open a roster at once: one member per pane, each tab named for its agent, running unattended. They coordinate over a shared bus and message each other directly.
• One sandbox for the team. Put the project and every agent in a single container instead of on your host — shared filesystem, one toolchain, and a real boundary around agents whose permission prompts you just turned off. Needs Docker or Podman and nothing else: no VS Code, no devcontainer CLI, no image to pull. The project mounts at its own path, so git worktrees stay valid from both sides.
• A worktree per agent. Create, open, and remove git worktrees as keybinds or MCP verbs, each opening a tab at the checkout — one isolated branch per agent, with triggers for setup and teardown.
• Resume, don't restart. With session-restore on, agent panes relaunch resumed, reloading their own session, instead of dropping to a bare shell.
• MCP server. sinclair mcp exposes the running terminal to any MCP client, so an agent can run commands, read the screen, switch tabs, replay macros, and manage worktrees.

Claude Code and Codex join the mesh natively over MCP. Ollama runs through a bridge; anything else runs via --cmd. The mesh is a bundled sidecar, managed from Settings → AI — it never runs inside the terminal process.

→ Agent teams · Sandbox · Worktree agents · MCP · docs/relay.md

A terminal that earns the tab

None of the above matters unless the terminal underneath is one you'd use anyway.

• Fast. GPU rendering with a damage-aware frame path: unchanged rows are reused across frames instead of re-shaped, and a scroll re-homes the rows it already has rather than rebuilding the grid.
• Complete. Sixel and kitty graphics, the kitty keyboard protocol, OSC 8 hyperlinks, OSC 52 clipboard, synchronized output, bracketed paste, mouse reporting, shell integration with jump-to-prompt, and reflow on resize.
• A real workspace. Tabs, recursive splits, a command palette, incremental scrollback search, copy/vi mode, broadcast input, and session restore.
• Yours. One settings.json that reloads the instant you save — fonts, theme, padding, cursor, and keybindings all update live. The settings UI (⌘,) writes back to the same file, so the file stays the source of truth.
• Extensible without a build step. Plugins are plugin.toml manifests contributing commands, live side panels, webview surfaces, event triggers, and MCP tools your agents can call — sandboxed WASM components that reach only what you grant them.
• Recording built in. Capture a pane to an asciinema .cast (⌘⇧R), then export it to GIF or MP4 rendered through the app's own text system, with the same ligatures and box-drawing you see on screen.
• Plus 22 themes, OS tabs (a Debian, Alpine, Ubuntu, Fedora, or Arch userland as a container-backed tab), recorded macros, and buffer export.

→ Feature coverage and known gaps · How Sinclair compares to kitty, Alacritty, Ghostty, and WezTerm

Install

macOS

brew install --cask wess/packages/sinclair

Or download Sinclair.dmg from the releases page and drag it to Applications.

Linux

Builds for x86_64 and aarch64 are on the releases page:

# AppImage — self-contained
chmod +x Sinclair-*-x86_64.AppImage && ./Sinclair-*-x86_64.AppImage

# Debian / Ubuntu
sudo apt install ./sinclair_*_amd64.deb

# Tarball
tar xzf sinclair-*-linux-x86_64.tar.gz

Sinclair draws its own window controls, so it needs a compositor with client-side decoration support (Wayland or X11).

From source

cargo run -p app --release

Configure

Press ⌘, for the settings window, or edit ~/.config/sinclair/settings.json directly — it's JSON with comments, it reloads the moment you save, and it only lists what you change.

{
  "font-family": ["JetBrains Mono", "Apple Color Emoji"],
  "font-size": 14,
  "theme": "catppuccin-mocha",
  "cursor-style": "bar",
  "scrollback-limit": 100000,
  "shell-integration": true,   // OSC 133/7 prompt-jump + cwd inheritance

  // AI — opt-in, also under Settings → AI
  "ai-enabled": true,
  "relay-enabled": true,
  "sandbox-enabled": false,    // run this project + its agents in one container

  "keybind": ["cmd+shift+c=copy_to_clipboard"]
}

A bad value becomes a friendly diagnostic on launch, falls back to its default, and never stops the rest of your settings from loading.

→ Every config key · Keybindings · Themes · Plugin tutorial

Documentation

The full documentation site covers install, configuration, keybindings, themes, plugins, and hands-on tutorials from your first split to a parallel agent team. In this repo:

docs/relay.md	The agent mesh: roles, teams, CLI, MCP tools
docs/sandbox.md	The shared project container
docs/compare.md	Sinclair vs. kitty, Alacritty, Ghostty, WezTerm
docs/parity.md	Terminal feature coverage and gaps
docs/libsinclair.md	Embedding the terminal in your own gpui app
docs/roadmap.md	Built vs. planned
docs/release.md	Signing, notarization, cutting releases

License

Apache License 2.0 · ♥ Sponsor this project