zero: a terminal coding agent with a security sandbox and support for 25+ LLM providers

What it solves

Zero is a terminal-based AI coding agent that allows developers to automate software engineering tasks directly from the command line while maintaining full control over the model used and the security of their local environment.

How it works

Zero operates as a CLI tool that can inspect repositories, edit files, and execute shell commands. It supports a wide array of LLM providers (including OpenAI, Anthropic, Gemini, and local options like Ollama) and implements a strict permission and sandbox policy to gate risky actions like network access or writing to files outside the workspace. It can be used interactively via a Terminal User Interface (TUI) or headlessly via zero exec for scripting and CI/CD integration.

Who it’s for

It is designed for developers who want a powerful AI coding assistant that is local-first, extensible, and provider-agnostic, avoiding lock-in to a specific AI service.

Highlights

• Provider Agnostic: Supports over 25 providers, including cloud APIs and local models.
• Security-First: Uses a sandbox policy and permission gating for file writes and shell commands.
• Local Context: Sessions are stored locally on disk and are searchable, resumable, and forkable.
• Extensible: Supports Model Context Protocol (MCP) servers, custom plugins, lifecycle hooks, and specialist subagents.
• Flexible Interface: Offers both a feature-rich TUI and a scriptable headless mode with JSON I/O.
• Customizable Instructions: Allows project-specific and personal guidance via ZERO.md or AGENTS.md files.