Observe: Know what's running
Every tab knows what's running inside it. Chau7 detects AI agents, shows their status, displays the repo name and branch, and tracks child processes. You see context, not just a shell prompt.
Free & Open Source
Every Model. One UI.
Named After a Sock
Chau7 is a macOS app that lets you run, compare, and steer coding agents across models. Locally.
A single terminal for all your AI agents.
Built for running multiple agents, not retrofitted for it.
Welcome to the shiny world of Agentic Engineering.
7+
Agent CLIs
1 UI
All of Them
1
Sock
$0
Open Source
You're running five agents. How many can you actually see?
You used to have one terminal, one shell, one train of thought. Now you have Claude in tab 1, Codex in tab 2, Gemini in tab 3, and no idea which one just ran rm -rf. Agentic engineering changed what a terminal needs to do. Most terminals didn't get the memo.
Chau7 is built for this: observe every agent, compare what they cost, intervene when they go sideways, and trust that nothing leaves your Mac.
Compare: See what they cost
Tokens, latency, and dollars per call, per session, per model. All tracked locally. Context Token Optimization strips terminal noise before your AI reads it. ~40% fewer tokens, lower bills.
Intervene: Step in when it matters
26 MCP tools let agents drive the terminal programmatically. Approval gates, dangerous command guards, and iOS Remote let you step in before things go wrong. From your desk or your Lock Screen.
Trust: Private by architecture
Zero network requests to any Chau7 server. All telemetry stays on your Mac. No account required. Built with Swift, Rust, and Metal. Open source. The architecture is the privacy policy.
Four jobs Chau7 does for you
01
Observe
Know what every agent is doing before you have to ask. Chau7 auto-detects running agents, brands tabs with their colors, and surfaces status, repos, and child processes.
You stop juggling context. Chau7 structures it for you.
See what they're doing →02
Compare
See what they cost before the invoice tells you. Tokens, latency, and dollars per call, per session, per model. Context Token Optimization saves ~40% on tokens.
Numbers instead of vibes.
Check the bill →03
Intervene
26 MCP tools let your agents drive the terminal. Approval gates, dangerous command guards, and tab limits keep you in control.
Step in from your desk or your Lock Screen.
Take the reins →04
Trust
Zero network requests to any Chau7 server. All telemetry stays on your Mac. No account, no analytics, no cloud.
Private because of how it's built, not what we promise.
See the architecture →Chau7 in action
AI Integration
Less noise for you. Fewer tokens for them.
Chau7 sits between you and your agents. It reduces cognitive load for the developer and reduces token waste for the agent. One window, every model, full visibility.
- Chau7 detects and identifies AI agents automatically.
- Every tab surfaces what matters: which agent is running, its live status, child processes, and previous session IDs. Quit and relaunch. Your tabs and AI sessions come back exactly where you left them.
- Context Token Optimization reduces token usage by ~40%.
- Context Token Optimization strips ANSI escape sequences, progress bars, and terminal noise before your AI reads the output. Cleaner context in, better output out, lower bill at the end.
- You finally see what your AI spends.
- Tokens, latency, and dollars per call, per session, per model. All tracked locally. No more waiting for the invoice to find out what happened.
Works with
Claude Code
Codex
Gemini CLI
ChatGPT
Copilot
Aider
Cursor
Missing yours? We'll add it.
Local MCP Server
20 MCP tools. Local Unix socket. Zero config.
Your agents talk to Chau7 through a local MCP server on a Unix socket. They can create tabs, run commands, read output, and query telemetry. No network, no cloud, no API keys. Auto-registers in Claude Code, Cursor, Windsurf, and Codex on every launch.
tab_exec call + response
tool: tab_exec
args: {
"tab_id": "C1C4AB49-DA22-46F6",
"command": "npm test"
}
result: {
"status": "executed",
"prompt_ready": true,
"pid": 48291
}Not Electron. Not a wrapper. Not sorry.
Running multiple agents means processing a lot of output, fast, without sending any of it somewhere you didn't ask for. Here's how.
Chau7 is a native macOS application built with Swift, Rust, and Metal.
The frontend uses Swift and AppKit for native macOS windowing, tabs, and menus.
The parsing backend uses Rust with SIMD-accelerated ANSI escape sequence parsing and a lock-free ring buffer between the PTY reader and the parser.
The renderer uses Apple Metal for GPU-accelerated glyph compositing and dirty-region redraws.
Privacy-first by architecture, not by promise. Zero network requests to any Chau7 server. Your data never leaves your Mac.
We did all of this for an app named after a sock. It's fine. We're fine. Mostly.
Browse all 178 features →iOS Remote
Your agent needed approval. You were at lunch.
Chau7 Remote lets you watch, approve, and redirect your agents from your iPhone. End-to-end encrypted with Curve25519 + ChaChaPoly1305. The relay never sees a byte.
1
Pair once
6-digit code. Curve25519 key exchange. Your devices remember each other by public key fingerprint.
2
Encrypted relay
ChaChaPoly1305 end-to-end encryption. The Cloudflare relay forwards opaque ciphertext it cannot read.
3
Approve from anywhere
View agent output, send commands, switch tabs, approve dangerous actions. Live Activities on your Lock Screen.