agentty vs Claude Code

Short version: agentty is a drop-in claude-code alternative that targets the same workflow — a coding agent in your terminal — but ships as a single native C++26 binary instead of a Node.js app, starts in under a millisecond, sandboxes every shell call by default, and is bring-your-own-model: run it with an API key, a local Ollama model, or your existing Claude Pro/Max OAuth. It works with any model (Claude, GPT, Groq, OpenRouter, Together, Cerebras, or local Ollama), not Claude only.

If you already like Claude Code's workflow but want it faster, dependency-free, model-agnostic, and sandboxed, agentty is built for you.

At a glance

Where agentty is different

It's a native binary, not a Node app

Claude Code is distributed as an npm package and boots a Node.js runtime on every invocation. agentty is a single statically-linked C++26 executable (13.6 MB) — curl | sh and you have one file with no runtime dependencies. Cold start is under a millisecond, and the TUI never pauses for garbage collection mid-stream. No version drift between machines, no npm install, no node_modules.

Any model, bring your own key

agentty is model-agnostic — run it with an API key for OpenAI, Groq, OpenRouter, Together, Cerebras, or Anthropic, a fully local Ollama model (no key, no cloud), or your existing Claude Pro/Max OAuth. Switch backends live in-app with ^P, or pass --provider. See Providers & Models. If you want Claude Code's ergonomics with a local, zero-API-cost model, that's a one-flag change.

Sandboxed by default, not as an afterthought

Every shell and build command agentty runs is wrapped in an OS-native sandbox — bwrap on Linux, sandbox-exec on macOS. Your workspace is read-write, system libraries are read-only, and ~/.ssh, /etc, and other projects are blocked. It's the default, not an opt-in. See Sandboxing.

Runs on air-gapped hosts

agentty airgap user@host relays traffic from your laptop to a machine with no direct internet, over SOCKS5-over-SSH with TLS pinned end-to-end. One command. See the air-gap guide.

Lives inside your editor over ACP

agentty acp runs the exact same engine as the TUI as an Agent Client Protocol agent inside Zed — streaming responses, inline diffs, native permission prompts, session reload — over JSON-RPC on stdio. Any ACP client works.

Where Claude Code is still ahead

Being honest: Claude Code is a mature, first-party Anthropic product with a large team behind it. It has broader IDE-extension coverage, a longer track record, and first-day support for new Anthropic features. agentty is pre-1.0 and moving fast — the core loop, tools, streaming, auth, and persistence all work and get daily testing, but treat it as a capable beta. If you want the officially-supported Anthropic experience above all else, use Claude Code.

When to choose agentty

Choose agentty if you want:

• A single-binary coding agent with zero runtime dependencies.
• Sub-millisecond startup and a TUI that never GC-pauses.
• Your choice of model — Claude, GPT, or a local Ollama model — behind one client.
• Shell calls sandboxed by default.
• To drive an agent on an air-gapped host over SSH.
• An open-source (MIT) tool you can read, fork, and ship.

Try it in one line

curl -fsSL https://agentty.org/install.sh | sh
cd your-project
agentty

First launch opens auth — paste an API key (any provider), use a local Ollama model with no key, or sign in with your Claude Pro/Max subscription. See the Quick Start or the full Installation guide.