zero

Zero is a coding assistant that runs in your terminal and takes instructions in plain English to read, edit, and run code in a project. Unlike cloud-hosted AI coding tools where sessions and data pass through a company's servers, Zero keeps everything local: your conversations are stored on your own disk, never uploaded as telemetry, and can be picked up again in a later session from where you left off.

One of the main design choices is that Zero does not lock you into a single AI model. You can connect it to over 25 different providers, including OpenAI, Anthropic, Gemini, Groq, DeepSeek, Mistral, and others. Local model runners like Ollama and LM Studio also work, so you can run entirely offline if you prefer. Switching the model is a single command or a menu selection.

When you open Zero it launches a terminal interface with a text area for your request, a sidebar that shows what the assistant is planning and doing, and controls for adjusting settings mid-session. You can ask it to fix a failing test, refactor a file, explain a piece of code, or work through a larger task step by step. For scripted use, a separate command called zero exec runs a single task without the interface, accepts and produces structured data formats, and exits with a meaningful status code so it can fit into automated workflows.

Zero has a layered permission system. Reading files is allowed by default. Writing files, running shell commands, and accessing the network require either your approval or a permission level you set upfront. If you want the assistant to also write to a folder outside the current project, you grant that explicitly with a flag or a slash command. This design is meant to make it clear what the assistant is doing and limit unintended changes.

The tool is written in Go, distributed as a single binary, and installs via npm, a curl script, or from source. It runs on Linux, macOS, and Windows.