The IDE that understands
AI agents

Run parallel agents on isolated branches with full observability.
Diffs, PRs, CI status, usage dashboards — one workspace, zero context loss.

10 AI agents detected natively

Complete workspace — editor, Git, diffs, PRs

AI-native, not AI-assisted

Download View on GitHub Nightly Builds

Install via terminal

Homebrew (macOS)

brew install sstraus/tap/tuicommander

Shell (macOS & Linux — auto-detects your system)

curl -fsSL https://tuicommander.com/install.sh | sh

PowerShell

irm https://tuicommander.com/install.ps1 | iex

All downloads & release notes →

TUICommander — multiple agents, split panes, sidebar with repos

Watch on YouTube →

macOS · Linux · Windows

Git worktrees, fully managed

"git worktree add ../feature-x origin/main && cd ../feature-x — every single time. And I have 8 worktrees open. Which one has the PR approved?"

Click a branch in the sidebar. TUICommander auto-creates the worktree, opens a terminal inside it, and cleans up when the branch is merged. The Worktree Manager gives you a bird's-eye view across all repos.

Automatic worktree creation on branch click, per-branch isolation
Warm worktrees — node_modules, target, .venv and every other ignored build directory arrive copy-on-write, so the first build is incremental and the disk cost is near zero (30 GB in ~38 s here)
Worktree Manager panel — PR status, dirty file counts, last commit at a glance
Post-merge cleanup dialog — delete branch, archive worktree in one click
Terminal auto-follows your cd — switch into a worktree and the tab reassigns
Configurable base branch, setup scripts, auto-fetch intervals

  WORKTREE MANAGER           PR    FILES

  myapp
    main                     —      —
    feature/auth       Open   +3
    bugfix/session     Merged  0
    refactor/api       Draft  +7

  [Clean up merged]  [Fetch all]
          

Observe your AI agents

"Your agent hit a rate limit 10 minutes ago. You didn't notice."

TUICommander understands what your agents are doing. Rate limits, questions, errors, and idle states surface automatically — no hunting through terminal windows.

Detect active tasks, waiting input, rate limits, and idle agents
Provider-specific countdown timers for rate limits
Notification sounds and keyboard overlay when agents need input
Activity dashboard showing all sessions at a glance

Claude Usage Dashboard — real-time rate limit bars, 7-day usage chart, 52-week contribution heatmap, per-project cost insights. Know exactly where your tokens go.

  CLAUDE USAGE DASHBOARD

  Rate Limits
  Opus    ████████░░  78%  resets 2:41
  Sonnet  ██░░░░░░░░  18%

  7-Day Usage
  Mon ██████  $4.20
  Tue ████    $2.80
  Wed ████████  $5.60
  Thu ██      $1.40

  52-Week Heatmap
  ░░▓▓██▓▓░░▓▓████▓▓░░
          

Progress — what happened while you were away

"I stepped away for two hours. What did the agents actually do?"

Every project gets an append-only journal. Agents write done when the work they announced is finished and blocked when they cannot continue without you — and TUICommander records the intent: each agent announced, so even an agent that never reports leaves a trail.

One compact MCP progress tool — kind, one sentence, an optional step label
Toolbar bell counts the entries that arrived since your last visit
Blocked entries in red, blocked-only filter, per-entry deletion — no pause, no clear, no correction
Stored in TUICommander's own database: nothing is written inside your repositories
A worktree files under its parent repository, so one project shows one history
One global switch plus a per-agent override turn collection off

  PROGRESS — tuicommander

  done     Route parity gate is green    Release prep
  blocked  Needs the signing password    Notarization
  intent   auditing the docs coverage
  ──────────────── last visit ────────────────
  done     WebSocket compression ack     Remote
  done     Dropped request no longer      AI Chat
           leaks a live ego
          

Built-in AI Chat & autonomous agent beta

"I want to ask a quick question about this terminal output without copy-pasting into another tool."

A conversational AI companion that sees your terminal as you see it. Ask about errors, get code suggestions, or let the autonomous agent take the wheel and drive your terminal directly.

Multi-provider: Ollama (local, free), Anthropic, OpenAI, OpenRouter, or any compatible endpoint
Streaming markdown with syntax highlighting and "Run this" action back to the terminal
Autonomous AI Agent (ReAct loop) with 12 tools: read screen, send input, edit files, search code, run commands
Session knowledge: the agent learns from your terminal history — commands, errors, fix patterns
Live cost tracking: prompt/completion tokens and estimated cost per turn
Per-terminal chat state, conversation history, detachable panel for multi-monitor

  AI CHAT (Ollama / local)

  You:  Why did this test fail?

  AI:   The test expects `auth.refresh()`
       to return a new token, but your mock
       returns `undefined`. Fix:

       mock.returnValue({ token: "abc" })

       [Run this]  [Copy]

  AGENT MODE (autonomous)
  read_screen OK → edit_file OK → run_command OK
  Cost: $0.0042  Cache: 87%
          

See what your agents changed

"Check the diff in VS Code. Check PR status on GitHub. Check CI in another tab."

The feedback loop happens in the same window — diffs, PR status, CI results, code editing, and automatic CI failure recovery.

Diff panel — working tree or last 5 commits, click for inline diff
PR status badges (open/merged/draft), CI rings, review state, merge & approve from the app
Multiple GitHub accounts — several github.com logins plus GitHub Enterprise Server (PAT), with per-repo account bindings and isolated per-account polling
GitHub Issues panel — filter by assigned/created/mentioned, labels with colors, close/reopen
Built-in CodeMirror 6 editor with syntax highlighting and disk conflict detection
CI Auto-Heal — fetches failure logs and injects them into your agent for automatic fix

  CHANGED FILES              +/-
  M src/auth/jwt.ts          +42 -18
  M src/auth/middleware.ts    +15 -3
  A src/auth/refresh.ts      +67
  D src/auth/sessions.ts     -89

  PR #247  Open  CI ●  Review: Approved
          

Control from anywhere

"My agents are running on my desktop. I'm on the couch with my phone."

Mobile companion PWA with live agent status, suggest chips, and question banners. Access via LAN, Tailscale with auto-provisioned HTTPS, or an E2E-encrypted cloud relay — no VPN or port forwarding needed.

QR code scan for instant LAN connection
Tailscale integration with automatic HTTPS certificate provisioning
Cloud relay — AES-256-GCM encrypted, zero-knowledge server
Answer agent questions with one tap, track rate limits live

tuic-remote headless daemon (beta) — run the backend where the code is: a Linux build host, a server, any machine with no desktop environment. Pure Rust binary, no Tauri and no GUI, serving the same HTTP/WebSocket API. Point a desktop TUICommander at it and its repositories, terminals and agents behave like local ones.

Linux x64/ARM64, macOS ARM and Windows x64 — shipped with tuic-bridge, which gives the agents on that machine the full tuicommander MCP tool surface
Every TCP request is authenticated: no loopback bypass, so an SSH tunnel does not make the daemon local
--instance <id> for an isolated configuration directory and keyring vault
Default port 9877, optional manual TLS, graceful SIGINT/SIGTERM shutdown

  MOBILE COMPANION

  feature/auth     Working...
  bugfix/session  ? Confirm (Y/n)  [Answer]
  refactor/api    Rate limited (2:41)

  Connected via Tailscale (HTTPS)

  HEADLESS DAEMON (build host)

  $ ./tuic-remote --instance build-host
  listening on 0.0.0.0:9877
  MCP bridge ready · every request authenticated
          

MCP Proxy Hub + Plugin system

Aggregate all your MCP servers into a single endpoint — every agent connects once and gets access to every upstream tool. Plus an Obsidian-style plugin system with hot reload and a community registry.

MCP Proxy Hub — single endpoint for all agents, auto-configures Claude Code, Cursor, Windsurf, VS Code, Zed, Amp, Gemini
Add or remove MCP servers without restarting anything
Circuit breakers, health checks, credential management, hot-reload
TUIC SDK v1.0 — tuic.activeRepo, tuic.toast, tuic.onRepoChange and more
Plugin system with 5 capability tiers, terminal watchers, and regex triggers
Community registry — browse and install with one click

Browse the plugin registry

// my-plugin/index.js — TUIC SDK v1.0
export function activate(tuic) {
  const repo = tuic.activeRepo;
  tuic.onRepoChange(repo => {
    tuic.toast(`Switched to ${repo.name}`);
  });
  tuic.registerWatcher({
    pattern: /error|fail/i,
    onMatch(line) {
      tuic.toast("Build failed!", "error");
    }
  });
}
          

Automate repetitive workflows

29 built-in Smart Prompts turn common tasks into one-click operations. Context variables like branch, diff, and PR data are resolved automatically — just click and go.

Smart Commit, Review Changes, Create PR, Fix CI — all one click
35 context variables auto-resolved from git, GitHub, terminal, and file context
Inject mode (PTY write), headless mode (one-shot subprocess), or shell script mode (direct execution, no AI needed, 60s timeout)
Shell scripts get the same context variables — automate git, deployment, and CI without an agent
Variable Input Dialog for any values that need manual input
Create your own prompts with the same variable system

// One click does all this:
// 1. Resolves {branch}, {base_branch}, {commit_log}
// 2. Injects into agent terminal:

Create a GitHub pull request for branch
feat/smart-prompts targeting main.
Generate a descriptive title and description
based on the commits:

a1b2c3 feat: add variable dialog
d4e5f6 feat: status banner
g7h8i9 fix: palette naming

Use `gh pr create`.
          

How it compares

TUICommander is an AI-native IDE built on a modern terminal.

Capability	Ghostty / Kitty	Cursor IDE	Claude Desktop	TUICommander
Terminal sessions	Yes	Yes	No	Yes (50)
AI coding agents	No	Built-in	Built-in	Any agent (10 detected)
Parallel agents	No	Limited	No	Unlimited
Git worktree orchestration	No	No	No	Automatic
Agent observability	No	No	No	Real-time
Global Workspace	No	Yes	No	Multi-repo
Usage dashboard	No	No	Basic	Full (heatmap, per-project)
Remote access (phone/tablet)	No	No	Mobile app	PWA + E2E relay
Voice dictation	No	Extension	Built-in	Local Whisper
MCP Proxy Hub	No	No	No	Built-in
Plugin system	No	Extensions	No	Hot reload + SDK
GitHub Issues & PR management	No	Extension	No	Built-in
Multi-account & Enterprise GitHub	No	No	No	github.com + GHE
Built-in AI Chat	No	Built-in	Built-in	Multi-provider (beta)
CI Auto-Heal	No	No	No	Built-in

And a full-featured terminal

Everything you'd expect from a modern terminal, plus a few extras.

50 sessions with independent zoom (8-32px)

Find in terminal — regex, case-sensitive, match nav

Session persistence — survives restarts

Branch panel — checkout, merge, rebase, push, pull

Repository groups — named, colored, collapsible

IDE launchers — VS Code, Cursor, Zed, Xcode

Command palette — fuzzy search all actions

Detachable tabs — float any terminal into its own window

Split panes — up to 4 panes, drag-resize, per-branch layout

11 bundled fonts — JetBrains Mono, Fira Code, and more

Claude usage dashboard — weekly limits, heatmaps

Smart Prompts Library — 29 built-in AI automation prompts + custom templates with auto-resolved context variables

Custom keybindings — rebind shortcuts, chord support, conflict detection

Tab status dots — 6 visual states: idle, running, unseen, awaiting, error

Post-merge cleanup — one-click branch delete, worktree archive

CWD tracking — tabs auto-follow when you cd into a worktree

GitHub Issues & PRs — unified panel with issue filters, labels, close/reopen, PR merge & approve

GitHub OAuth — Device Flow login, OS keyring storage, no PAT needed

Multi-account GitHub — several github.com logins + GitHub Enterprise (PAT), per-repo bindings, isolated polling

CI Auto-Heal — fetches failure logs, injects into agent for automatic fix

File browser — tree/flat view, content search with regex, case-sensitive, whole-word

Markdown notes — interactive GFM checkboxes, live preview, write back to disk

Native drag and drop — drop files and folders into the terminal

Open / New file — Cmd+O and Cmd+N open the built-in editor directly

Font size inheritance — terminal inherits editor font size, or set independently

Cross-repo knowledge — mdkb-powered search across all repos via MCP

Worktree agent bridge — spawn AI subagents in isolated worktrees via MCP

Frequently Asked Questions

Is it free?

Yes. TUICommander is Apache 2.0 licensed, open source, and free forever. No subscriptions, no feature gates.

Does it work on Windows and Linux?

Yes. TUICommander is cross-platform — macOS, Windows, and Linux. macOS builds are signed and notarized for one-click install.

Is this an IDE?

Yes. TUICommander is an AI-native IDE — the IDE that understands AI agents. It includes a code editor, file browser, diff views, Git panel, PR management, CI Auto-Heal, and a usage dashboard. Unlike traditional IDEs that bolt on AI features, TUICommander is designed from the ground up for working with multiple parallel agents as the primary workflow, with full observability into what each agent is doing.

Can I use it without AI agents?

Yes. TUICommander is a powerful workspace on its own — 50 sessions, split panes, git worktrees, code editor, diff views, find-in-terminal, session persistence, and more. Agent awareness is a core capability, not a requirement.

Can I monitor agents from my phone?

Yes. TUICommander includes a mobile companion web app you can add to your home screen. See which agents need input, answer questions with one tap, track rate limits, and watch live output — all from your phone or tablet.

What happens when I restart the app?

Everything comes back. Terminals, tabs, split layouts, and scroll history are persisted across restarts. Agent sessions are detected automatically — you get a clickable banner to resume exactly where you left off.

How does voice dictation work?

Press a hotkey and talk. Your speech is transcribed locally and injected into the active terminal. Give detailed, nuanced prompts to AI agents in seconds instead of spending minutes typing. GPU-accelerated on macOS, no cloud service needed.

What's the MCP Proxy Hub?

Connect your MCP servers to TUICommander once, and every AI agent you run gets access to all of them automatically. No per-agent configuration. Includes circuit breakers, health checks, and hot-reload — add or remove servers without restarting anything.

Can I extend it with plugins?

Yes. TUICommander has an Obsidian-style plugin system with hot reload and a community registry. Plugins can watch terminal output, add status bar widgets, create panels, and respond to agent events. See the Plugin Authoring Guide.

What is the TUIC SDK?

The TUIC SDK (v1.0) is the plugin API that gives your extensions access to the TUICommander runtime. It exposes tuic.activeRepo, tuic.toast(), tuic.onRepoChange(), terminal watchers, and more. Plugins are plain JavaScript modules with hot reload — no build step required.

How does the cloud relay work?

The relay is a lightweight WebSocket server that forwards encrypted messages between your desktop and phone. All traffic is end-to-end encrypted with AES-256-GCM — the relay server is zero-knowledge and never sees your data. The encryption key is derived from your token via HKDF and embedded in the QR code. You can self-host the relay or use the default at relay.tuicommander.com.

Is my data sent anywhere?

No. Zero telemetry, zero analytics, zero cloud dependencies. Everything runs locally on your machine. The only optional network calls are to GitHub's API (for PR badges), the plugin registry, and the cloud relay (if you enable it — all traffic is E2E encrypted).

How does the AI Chat work?

AI Chat (beta) is a conversational AI companion built into TUICommander. It sees your terminal output and can answer questions, suggest fixes, or take autonomous action via the AI Agent loop. It supports multiple providers: Ollama (runs locally for free), Anthropic, OpenAI, OpenRouter, or any OpenAI-compatible endpoint. Each terminal gets its own independent chat state. API keys are stored in the OS keyring. The autonomous agent has 12 tools including screen reading, file editing, code search, and command execution — with safety checks and user approval for destructive actions.

Does it support Claude Code Agent Teams?

Yes. TUICommander automatically enables Claude Code's experimental Agent Teams feature. Tell Claude to "create an agent team" and each teammate spawns as a separate terminal tab. Teammates share a task list, message each other directly, and coordinate autonomously — all visible in your sidebar. No tmux or iTerm2 required.

The IDE that understands
AI agents

Works with your favorite AI coding agents

The problem

Run many AI sessions in parallel

Git worktrees, fully managed

Observe your AI agents

Progress — what happened while you were away

Built-in AI Chat & autonomous agent beta

See what your agents changed

Global Workspace & Multi-Monitor

Control from anywhere

MCP Proxy Hub + Plugin system

Automate repetitive workflows

Customize your workflow

Talk to your agents

How it compares

And a full-featured terminal

Frequently Asked Questions

Ready to work with AI agents without losing context?

The IDE that understandsAI agents

Works with your favorite AI coding agents

The problem

Run many AI sessions in parallel

Git worktrees, fully managed

Observe your AI agents

Progress — what happened while you were away

Built-in AI Chat & autonomous agent beta

See what your agents changed

Global Workspace & Multi-Monitor

Control from anywhere

MCP Proxy Hub + Plugin system

Automate repetitive workflows

Customize your workflow

Talk to your agents

How it compares

And a full-featured terminal

Frequently Asked Questions

Ready to work with AI agents without losing context?

The IDE that understands
AI agents