If you like Claude Code's terminal-first workflow but want something open source, faster, and not locked to a single provider, agentty is built for exactly that.
Why people look for a Claude Code alternative
- They want an open-source agent they can read, self-host, and contribute to.
- They want to use providers other than Anthropic — OpenAI, Groq, OpenRouter, Cerebras, or a fully local Ollama model.
- They don't want a Node.js runtime and
npm installon every machine. - They want stronger isolation — a sandbox around shell and build commands by default.
How agentty compares
agentty keeps the part you like — a fast, keyboard-driven terminal agent — and changes the parts that matter for power users:
- Single static binary. 16.7 MB, ~3 ms cold start, no Node, no Python, no Electron.
- Bring your own model. Sign in with your existing Claude Pro/Max, or use API keys for OpenAI, Groq, OpenRouter, Cerebras, or point it at local Ollama.
- Sandboxed by default. Shell and build calls run isolated; air-gap a whole session over SSH with one command.
- Runs in Zed over ACP, and speaks MCP.
- Open source (MIT).
Install in one line
curl -fsSL https://agentty.org/install.sh | sh
No Node, no Python — just a single static binary.
Is it really a drop-in replacement?
For the core loop — chat, read/edit files, run commands, retrieve from your repo — yes. If you depend on a Claude-Code-only feature the day it ships, keep Claude Code for that. For everyone who wants an open, fast, multi-provider agent, agentty is the alternative to try first.