Attyx
A GPU-accelerated terminal that tracks, coordinates, and controls AI coding agents through native MCP, IPC, and CLI tooling.
Attyx earns a place on your machine if you already run more than one coding agent at a time. The per-pane status dots and the ⌘⇧A dashboard that reports tokens, cost, and context window per agent solve a real problem: with four Claude Code or Codex panes open you genuinely lose track of which one is blocked on input and which is burning budget. The MCP server and attyx CLI go further than any mainstream terminal, letting one agent spawn and drive another. If you run a single agent occasionally, a mature terminal like iTerm2, Kitty, or Alacritty plus tmux covers you, and Attyx's TOML-only configuration and macOS/Linux-only platform support may not be worth the switch. Adopt it for agent
Verified 6d ago · liveness 69/100 · cite: rightaichoice.com/tools/attyx
- Developers running several AI coding agents in parallel
- Agent workflow orchestrators coordinating multiple agents
- Terminal power users who want scriptable control over panes
- Teams standardized on Claude Code, Codex, opencode, or pi.dev
- Windows users (macOS and Linux only)
- Users who want GUI-based settings instead of a TOML file
- Anyone whose workflow depends on a large third-party plugin marketplace
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Attyx if you run one agent at a time or need a Windows build — its value is orchestration across multiple Claude Code, Codex, opencode, or pi.dev panes, and it is documented for macOS and Linux only.
Attyx is MIT-licensed and installs through the semos-labs Homebrew tap (brew install semos-labs/tap/attyx), so there is no license line item to compare against paid terminals or agent dashboards. The cost that actually lands is the agents it supervises: the dashboard surfaces per-agent spend — the vendor's own example shows $0.42 for one pane and a $0.73 session total — which is inference billed by your model provider, not by Attyx.
In short
Attyx — A GPU-accelerated terminal that tracks, coordinates, and controls AI coding agents through native MCP, IPC, and CLI tooling. Best for Developers running several AI coding agents in parallel, Agent workflow orchestrators coordinating multiple agents, Terminal power users who want scriptable control over panes. Free to use.
What's new in Attyx
Checked 6 days agoAcross the latest 1 update: 1 changelog entry.
What people actually say about Attyx — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
14 mentions across 4 sources (Hacker News, YouTube, GitHub, Lemmy) · researched Aug 31, 2026.
Average across the 4 sources that answered — each source counts once, not each post.
- +GPU-accelerated rendering via Metal/OpenGL is fast and smooth.
- +Single binary under 5 MB — no Electron bloat, impressive.
- +Native MCP over stdio and loopback HTTP enables powerful automation.
- +Agent lifecycle status dots and dashboard are genuinely useful.
- +Daemon-backed sessions survive terminal closures.
- −Coredumps on exit are a dealbreaker for some users.
- −Fractional scaling issues plague Wayland with Niri.
- −Brand confusion with a solar company creates PR headaches.
- −Small community (239 stars) — limited support and resources.
- −Advanced features require a learning curve for newcomers.
- • None — completely free, but costs time to configure and troubleshoot early bugs
Viability Score
How well maintained and how widely used is Attyx? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: September 2026
How we score →Key Features
- Agent lifecycle status dots per pane (working, waiting on you, idle)
- Full-screen cross-session agent dashboard via ⌘⇧A
- Live per-agent tokens, elapsed time, context window, and estimated cost
- Native MCP server over stdio and loopback HTTP at 127.0.0.1:7333/mcp
- IPC layer over a Unix socket with stable per-pane IDs
- attyx CLI for split, send-keys, get-text, watch, and split close
- Claude Code skill installation via attyx skill install
- Claude and Codex JSON lifecycle hook integration
- opencode event-bus plugin and pi.dev event-bus extension
- GPU-accelerated rendering: Metal on macOS, OpenGL on Linux
- Single native binary under 5 MB, written in Zig 0.15
- Daemon-backed persistent sessions that survive closing the window
- Vertical and horizontal splits, tabs, and floating popups
- Fuzzy-searchable command palette (Cmd+Shift+P / Ctrl+Shift+P)
- 22 built-in themes with live preview (Catppuccin, Dracula, Gruvbox, Nord, Tokyo Night)
About Attyx
Attyx is a terminal emulator built specifically for running AI coding agents rather than just shells. It detects the lifecycle of Claude Code, Codex, opencode, and pi.dev — reading Claude and Codex through JSON lifecycle hooks and opencode and pi.dev through event-bus extensions — then paints a colored status dot on each pane and tab showing whether an agent is working, waiting on you, or idle. Pressing ⌘⇧A opens a full-screen dashboard (Attyx calls it an htop-style view) that lists every agent across every session with live state, elapsed time, tokens in and out, context-window usage, and estimated cost, and lets you focus, zoom, or close an agent without leaving the table. Underneath the agent layer it is a real terminal: vertical and horizontal splits, tabs, floating popups, a fuzzy command palette (⌘⇧P), regex scrollback search, vim-style visual mode, the Kitty graphics protocol for inline images, and 22 themes including Catppuccin, Dracula, Gruvbox, Nord, and Tokyo Night. It is written in Zig, GPU-rendered via Metal on macOS and OpenGL on Linux, and compiles to a single binary under 5 MB with no Electron or bundled browser. Sessions are daemon-backed, so you can close the window and reopen tomorrow with every pane where you left it. The part that distinguishes Attyx from a regular terminal is programmability: every pane has a stable ID, an IPC layer over a Unix socket exposes panes and keystrokes, the attyx CLI can split, send keys, and read output (as in id=$(attyx split v --cmd "claude -p 'run tests'") followed by attyx send-keys), and a native MCP server over stdio or a loopback HTTP endpoint (127.0.0.1:7333/mcp) exposes the terminal as typed tools to Claude Desktop, Claude Code, or any MCP client. An attyx skill install command drops a /attyx skill into Claude Code. Configuration is TOML at ~/.config/attyx/attyx.toml with hot reload. Attyx is MIT-licensed and installs via Homebrew on macOS (Apple Silicon and Intel) and Linux (x64 and ARM64); on Linux it installs as a desktop application. It is a desktop and CLI tool — there is no web or mobile version, and no Windows build is described in the documentation.
Behind the Verdict
Attyx has a clear thesis and executes it narrowly, which is the right approach for a terminal. The core observation is that a terminal emulator used to be judged on rendering speed and keybindings; once your panes are running Claude Code, Codex, opencode, or pi.dev, the thing you actually need to see is which agent is working, which is waiting on you, and what each one is costing. Attyx answers that with a per-pane colored dot and a cross-session dashboard listing state, elapsed time, input/output tokens, context-window fill (the sample shown reports values like 82k/200k and 55k/256k), and estimated dollar cost per agent with a session total. Strengths. The agent-awareness is sourced properly rather than guessed: Claude and Codex lifecycle come through JSON hooks and opencode and pi.dev through event-bus extensions, so the status is deterministic instead of scraped from screen text. The programmability layer is unusually deep for a terminal — stable pane IDs, a Unix-socket IPC layer, an attyx CLI with split/send-keys/get-text/watch subcommands, and a native MCP server over stdio or 127.0.0.1:7333/mcp, plus attyx skill install to hand Claude Code typed control of the terminal. The engineering footprint is small and honest: Zig 0.15, a single binary under 5 MB, Metal and OpenGL rendering, no Electron or Chromium. Standard terminal expectations are not sacrificed — Kitty graphics and keyboard protocols, OSC 8 hyperlinks, bracketed paste, IME support for CJK, synchronized output, DECSTBM scroll regions, reflow on resize, 24-bit true color, and a tmux-style daemon backing sessions. Weaknesses and fit. This is an operator console for people running agents, and the configuration story assumes you are comfortable editing attyx.toml — there is no GUI settings panel. Agent lifecycle detection is explicitly scoped to Claude Code, Codex, opencode, and pi.dev; an agent outside that list gets no status dot. GPU acceleration means Metal on macOS and OpenGL on Linux, and the documented platforms are macOS (Apple Silicon and Intel) and Linux (x64 and ARM64), with no Windows build described. Users migrating off iTerm2, tmux, or Kitty for their plugin ecosystems will find Attyx deliberately has no plugin marketplace — instead you get the CLI and MCP surface as the extension mechanism. Where it fits. Solo developers or small teams running several agents in parallel who want one screen showing what needs attention and what it costs. Where it doesn't: someone who wants a friendly general-purpose terminal, someone on Windows, or someone whose workflow depends on a large third-party plugin library rather than scripted control.
Researching Attyx? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Attyx actually fits — and what changes day-one when you adopt it.
You open a web pane, an api pane, an infra pane, and a test pane, each running Claude Code or Codex on a different part of an app, then hit ⌘⇧A to pull up the dashboard.
Outcome: The dashboard lists all four agents with state, elapsed time, tokens in and out, context-window fill, and estimated cost, and the status dot flags which pane flipped to waiting on you so you stop babysitting the three that are still working.
You script a coordinator that calls id=$(attyx split v --cmd "claude -p 'run tests'") to launch a worker in a new pane, polls it with attyx get-text -p $id until it prints a pass count, then nudges it with attyx send-keys -p $id "yes{Enter}" when it asks for confirmation.
Outcome: One agent drives several others through stable pane IDs over the Unix socket, and attyx split close -p $id tears each worker down when the run finishes.
You add Attyx to claude_desktop_config.json as an MCP server with command "attyx" and args ["mcp"], then ask Claude to open a pane and read what it prints.
Outcome: Claude gets the terminal as typed tools over stdio — panes, keystrokes, and output — so it drives the real running terminal instead of a sandbox, and the same tools are reachable from any MCP client at 127.0.0.1:7333/mcp.
Use Cases
- Watch several Claude Code or Codex panes at once and see which one is blocked waiting on input
- Track per-agent token burn, context-window fill, and estimated cost from a single dashboard
- Have one agent spawn a pane, run a command, and read the output of another via the attyx CLI
- Point Claude Desktop or any MCP client at attyx mcp to drive a real running terminal
- Keep long-running agent sessions alive across window closes with daemon-backed sessions
- Script terminal interaction over the Unix-socket IPC layer for CI-style automation
- Install the Claude Code /attyx skill so Claude gets typed control of panes and keystrokes
- Lay out agent and editor panes with splits, tabs, and floating popups on a GPU-rendered terminal
Models Under the Hood
as of 2026-09-29
Limitations
- Agent lifecycle status detection covers four agents — Claude Code, Codex, opencode, and pi.dev — so an agent outside that list does not get a per-pane status dot.
- GPU acceleration uses Metal on macOS and OpenGL on Linux, and documented platforms are macOS (Apple Silicon and Intel) and Linux (x64 and ARM64).
- Configuration lives in ~/.config/attyx/attyx.toml, and building from source requires Zig 0.15.2 or newer.
- It is a desktop and CLI application rather than a web or mobile one.
as of 2026-09-22
Verification history
We have re-verified Attyx 8 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 8 verification passes.
Free to cite with attribution — this page re-verifies continuously.
Where the pricing makes sense
The company stage and team size where Attyx's pricing actually pencils out — and where peers do it cheaper.
Attyx is MIT-licensed and installs through the semos-labs Homebrew tap (brew install semos-labs/tap/attyx), so there is no license line item to compare against paid terminals or agent dashboards. The cost that actually lands is the agents it supervises: the dashboard surfaces per-agent spend — the vendor's own example shows $0.42 for one pane and a $0.73 session total — which is inference billed by your model provider, not by Attyx.
Setup time & first value
How long it actually takes to get something useful out of Attyx — broken out by persona, not the marketing-page minute.
Install is a two-command Homebrew path (brew tap semos-labs/tap, then brew install attyx) and the docs note Linux users may need to log out and back in if the desktop entry does not appear — under five minutes to a working terminal. First value on a multi-agent workflow arrives as soon as you add the Claude Code skill with attyx skill install and wire the JSON hooks or event-bus extensions for
Switching to or from Attyx
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From iTerm2 or Alacritty: install via Homebrew, copy your key preferences into ~/.config/attyx/attyx.toml, and use ⌘⇧P to find the action you used to reach through a menu
- →From tmux: the daemon-backed session model replaces the tmux server, and native splits, tabs, and floating popups remove the need for a tmux.conf of split bindings
- →From Kitty: the Kitty graphics and keyboard protocols carry inline images and unambiguous key reporting over, and you drop the config file for attyx.toml with hot reload
- →From Terminal.app: Homebrew install, then set status-bar widgets for cwd and git branch to replace the title-bar habits you had
- ↗To tmux or Zellij: export your routine split and session layouts as shell aliases, then rebuild them as tmux panes once you no longer need the attyx CLI or MCP surface
- ↗To iTerm2 or Kitty: carry your themes over by translating the Attyx TOML colors into the target terminal's scheme format
- ↗To a script-only setup: replace the attyx split/get-text/send-keys calls with your terminal's own scripting interface, accepting the loss of stable pane IDs and typed MCP tools
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Attyx”, and we withheld 6: 6 could not be judged, because “Attyx” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Attyx.
Official links
Tools that pair well with Attyx
Common stack mates teams adopt alongside Attyx, with the specific reason each pairing earns its keep.
Warp
Build, run, and orchestrate fleets of AI coding agents across your SDLC with Warp Factories, Warp Terminal, and the Warp Agent CLI
Oh My Pi
Terminal-native AI coding agent with hash-anchored edits and 40+ model providers
GitHub Copilot
GitHub Copilot is an AI coding agent that completes code, reviews pull requests, and runs agents from issue to merge
Featured Head-to-Head Comparisons
Attyx vs Spider Cloud
Spider Cloud and Attyx serve entirely different layers of the AI stack. Spider Cloud is a data ingestion tool—feed it URLs and get structured content for LLMs. Attyx is an execution environment—run and orchestrate agents live. If you need fresh web data for your RAG pipeline or agent, go with Spider Cloud. If you're wrangling multiple coding agents in a terminal and want native MCP support, pick Attyx. They complement rather than compete, so don't force a choice unless your problem is specifically about web data acquisition.
Attyx vs Voyage Ai
Voyage AI and Attyx serve completely different niches. Voyage AI is best for enterprise teams building RAG pipelines needing high-accuracy, domain-specific embeddings with compliance (SOC 2, HIPAA); Attyx is a free, GPU-accelerated terminal for developers orchestrating AI coding agents like Claude Code. If you're embedding documents for retrieval, choose Voyage; if you're managing agent workflows in a terminal, choose Attyx.
Attyx vs Temporal Ai
If you need a battle-tested backend orchestrator for reliable, long-running AI workflows with automatic retries and state persistence, Temporal AI is the choice (trusted by OpenAI). If you're a developer running multiple AI coding agents locally and want a fast, scriptable terminal with native MCP support and agent-aware monitoring, Attyx is a compelling free tool. They complement rather than compete.
Alternatives to Attyx
View allWarp
Build, run, and orchestrate fleets of AI coding agents across your SDLC with Warp Factories, Warp Terminal, and the Warp Agent CLI
GitHub Copilot
GitHub Copilot is an AI coding agent that completes code, reviews pull requests, and runs agents from issue to merge
Frequently Asked Questions
Categories
Best-of guides
Used Attyx? Help shape our editorial sentiment research.