patriceckhart/zot

What it solves

Zot is a lightweight coding agent harness that allows users to interact with various LLMs to perform software development tasks. It simplifies the process of managing multiple AI providers, handling authentication (including subscription-based logins), and providing the agent with the necessary tools to read, write, and execute commands on a local filesystem.

How it works

Written in Go as a single static binary, Zot acts as an intermediary between the user and an LLM. It provides a set of built-in tools (read, write, edit, and bash) that the agent can use to interact with the codebase. It supports a wide array of AI providers via API keys or OAuth flows and allows for the customization of agent behavior through persistent instruction files (AGENTS.md and SYSTEM.md) and reusable skills. It can be run in several modes, including an interactive TUI, a print mode for scripts, and an RPC mode for embedding into other applications.

Who it’s for

It is designed for developers who want a flexible, provider-agnostic coding assistant that can be run from the command line, integrated into automation scripts, or embedded into other software tools.

Highlights

• Extensive Provider Support: Built-in support for over 30 providers including Anthropic, OpenAI, Google Gemini, DeepSeek, and local models via Ollama.
• Flexible Authentication: Supports API keys, external credential commands (e.g., via password managers), and subscription-based OAuth logins for services like Claude Pro and ChatGPT Plus.
• Customizable Instructions: Uses a hierarchical discovery system for AGENTS.md files to layer project-specific or global instructions on top of default prompts.
• Extensible Architecture: Supports extensions written in any language via subprocesses and JSON-RPC.
• Portable Agents: "Zotfiles" allow agents to be packaged and shared via local directories, .zot files, or public GitHub repositories.
• Multiple Interface Modes: Offers an interactive TUI, a simple print mode for CLI piping, and a long-lived RPC process for third-party integration.