ariel-frischer/arc-agent

Arc-agent is a command line tool written in Go that uses AI to produce structured system design documents for software projects. System design is the process of planning how a software system's parts, such as databases, APIs, services, and data flows, fit together before or during development. This tool automates that planning by running a multi stage workflow where each stage, including requirements, entities, API shape, high level architecture, and optional deep dives, is handled by a separate AI call, checked against a schema, and repaired automatically if the output is invalid. The final output is a workspace folder containing validated YAML files for each stage, a DESIGN.md document in plain Markdown, and a Mermaid diagram file representing the architecture. Mermaid is a text based format that renders into visual flowcharts and diagrams. The tool can also export Excalidraw format diagrams and render SVG or PNG images from them. Arc-agent supports multiple AI runners. It can call OpenCode, Codex, or Claude Code as an agent inside an isolated workspace, or it can call the OpenCode Go API directly for faster, more predictable structured output. An auto mode picks the appropriate runner based on which API keys and tools are available locally. A person would use this when starting the design of a new system, such as a URL shortener, a messaging platform, or any multi component service, and wants a structured, reviewable starting point instead of a blank page. The staged workflow is inspired by a system design interview preparation guide that focuses on practical requirements, trade-offs, and clear communication. The tool installs through a shell script and runs on macOS and Linux, with Windows users expected to use WSL. It is released under the MIT license.