l0ng-ai 2e6f9cab4f perf(protocol): put a frame's header and payload on the wire in one write (#713)
`write_frame` sent the length, the kind byte and the payload as three
separate `write_all` calls. The pane socket is a loopback `TcpStream` with
`TCP_NODELAY` set, so each of those was its own segment and its own wakeup
on the far side: the client woke from `read()` about twice per frame just
to reassemble a header it had already been told the shape of. The daemon
frames every ConPTY read, so under output that is tens of thousands of
extra syscalls a second on each end.

Write the 5-byte header and the payload as one `write_vectored` instead.
The bytes on the wire are identical, and a writer with no native vectored
write still terminates on the loop's fallback — it just costs what it
used to.

Measured on a Ryzen 9 9950X (32 threads, integrated Radeon), release
build, one pane in a 2182x1361 window, a generator emitting 20000
80-column lines/s (1.64 MB/s), three 25-second samples each. CPU is % of
one core; "reads/frame" is `TTY7_TRACE`'s client socket reads divided by
its frame count.

| | tty7-app (GUI) | daemon | conpty host | total | reads/frame |
|---|---|---|---|---|---|
| before | 22.0 | 3.8 | 2.3 | 28.1 | 1.90 |
| after | 17.7 | 3.1 | 2.5 | 23.3 | 0.93 |

Socket reads per frame halve, and CPU at a fixed output rate falls 17%.
Under an unbounded flood the pipeline instead gets faster — 35.7 MB/s
before, 47.0 MB/s after — so that case is not rate-matched and is not
claimed as a CPU win.

This does not on its own explain #713: at the rate that issue describes
(a held Return, ~30 lines/s) tty7 costs about 11% of one core here either
way. It is a real cost on the output path regardless.
2026-09-07 19:31:09 +08:00
2026-09-04 17:42:08 +08:00
2026-09-04 17:42:08 +08:00
2026-09-04 17:42:08 +08:00

tty7

tty7

A terminal workbench: persistent sessions, remote work, agents.

Pure Rust · GPU rendering on Zed's gpui · VT core from Alacritty


CI Version Platforms License Discord

English · 简体中文


tty7 with a tab sidebar of agent sessions across several repos, running Claude Code

Why

A background server owns your shells and panes — not the window. Everything below follows from that.

  • Performance — ~2× the throughput of Alacritty, Ghostty, or Kitty (benchmarks)
  • Persistent sessions — quit or reboot; your shells and supported agent sessions keep running, no tmux
  • Agent-aware — Claude Code, Codex & co.: status, notifications, and git context for every repo at once
  • Scriptable by agents — one agent opens a pane for another, hands off a task, waits, and reads the result, with or without the GUI running
  • Editor-grade input — suggestions, completion, highlighting, history search, with no plugin to install
  • Remote development — files, repos, panes, and git data stay on the remote machine, over a native SSH stack
  • Git beside the terminal — source control, diffs, and worktrees without leaving the window

Install

Native builds for macOS, Windows, and Linux on Releases:

macOS …-macos-arm64.dmg · …-x86_64.dmg drag into Applications
Windows …-setup.exe · portable ….zip
Linux …-x86_64.AppImage chmod +x and run — X11/Wayland libraries bundled

What's inside

Agent-aware per-pane detection (19 CLIs) · status dot · notifications · branch + diff · tray icon when input is needed · resume after reboot · tab sidebar grouped by repository
CLI + Skills bundled tty7 CLI · agent skill · run streams a command and exits with its code · split · send · wait --until free · capture
Editor-grade input ghost suggestions from history · explained tab completion · syntax highlighting · multi-line editing · click places the caret · ⌃ R fuzzy history
Window tabs & splits · ⌘ P palette · ⌘ F scrollback search · ⌘ J panel with process tree and listening ports · 13 themes, your own YAML, iTerm2 import · IME
Shell integration injected when a pane starts, nothing to install · prompt marks · working directory · exit codes · command-finished notifications · zsh, bash, fish, PowerShell, WSL, remote panes
Remote workspaces remote files, repos, changes, diffs, worktrees, tabs, and panes · reconnect from any client and continue where you left off
SSH native russh stack: profiles with keychain secrets · SFTP panel · port forwarding · jump hosts · one-time, unprivileged tty7-server install
Git panel follows the focused pane · stage, commit, amend, branch, push, stash · side-by-side or unified diffs · commit graph with cherry-pick, revert, and reset · a new worktree opens its own tab

Supported agents

Detection is free: brand avatar, branch + diff, tab title. Status takes one click under Settings → Agents to install that agent's hook, and brings the status dot, notifications, the tray icon, tty7 wait, and resume after a reboot. Fork needs both — the agent's own fork command, and the hook that tells tty7 which session to fork.

The full support matrix, all nineteen
Agent Detected Status · resume Fork
Claude Code
Codex
Grok
OpenCode
Oh My Pi
Droid
Qwen Code
Goose
Gemini
Copilot
Kimi Code
Pi
Aider
Amp
Cursor
Auggie
Hermes
Vibe
Antigravity

None of them are wrapped or proxied — the agent you start is the agent you get, in a normal PTY, with its own interface. An agent launched through a wrapper script can be mapped to one by name with agent_commands in config.json.

Documentation

Full documentation lives in docs/keyboard shortcuts · config.json · CLI reference. The agent-facing CLI interface is also documented in skills/tty7/SKILL.md.

Install the skill with:

npx skills add l0ng-ai/tty7    # install
npx skills update tty7         # update later

Benchmarks

Same machine, same day, same 155×40 grid — Apple M1 Pro, macOS 26.3.1, five-run averages (2026-07-04):

tty7 Alacritty Ghostty Kitty
Plaintext I/O — 11 MB cat (lower = better) 95 ms 239 ms 179 ms 185 ms
DOOM-fire frame rate (higher = better) 888 fps 485 fps 552 fps 617 fps
Cold-launch memory 116 MB¹ 105 MB 128 MB 130 MB

¹ GUI 105 MB + the persistent server 11 MB.

Methodology and one-command reproduction: scripts/bench/.


S
Description
A terminal workbench in pure Rust: shells, persistent sessions, SSH, coding agents. GPU-rendered on Zed's gpui, VT core from Alacritty.
Readme Apache-2.0
34 MiB
Languages
Rust 99.1%
Shell 0.5%
PowerShell 0.2%
Inno Setup 0.1%