What it does
agentty is a coding agent that runs in your terminal. It ships as one native binary with no Node or Python runtime, and the README quotes a cold start of a few milliseconds. You launch it inside a project and type tasks into a composer, with tools for reading, writing and editing files, running shell commands, searching and git. It is not tied to one model vendor. You can paste an API key, sign in with a Claude Pro or Max account, or point it at providers such as OpenAI, Groq, OpenRouter, DeepSeek, xAI, Gemini or a local Ollama model, and switch between them from inside the app. An optional Smart Mode lets a stronger model plan while cheaper models handle smaller subtasks. To keep prompts small it includes local retrieval that searches your docs, skills, saved memory and source code instead of sending the whole repository. Shell commands run in an operating-system sandbox by default, and an air-gapped mode relays traffic over SSH for machines without internet. It can also run a single prompt non-interactively for scripts and CI, work inside Zed over the Agent Client Protocol, and load Agent Skills from SKILL.md files.
Capabilities
- Single native binary with no runtime dependencies
- Works with many providers, including local Ollama models
- Sandboxed shell and build commands by default
- Local retrieval over docs, skills, memory and source code
- Smart Mode splits planning and routine work across models
- Non-interactive mode for scripts and CI, and Zed integration
When to use it
- You want a terminal coding agent that is not locked to one model provider
- You work on a machine without internet access and need an agent that runs over SSH
- You want to script one-off coding prompts in CI with machine-readable output
Alternatives to explore
- Claude Code
- Aider
- Cursor
