Maggy

Maggy is an open source engineering harness for AI coding assistants, released under the MIT license by the team behind Palgu. It layers a test enforced, quality gated workflow on top of Claude Code, Codex, Kimi and Gemini CLI. It is aimed at developers who want more structure and lower model cost from command line coding agents. Its main components are two parts in one repository. Claude Bootstrap is an installable configuration pack of 67 skills, hooks, rules and templates. The optional Maggy server is a local FastAPI service with a web dashboard that adds routing, skill protocols, a code graph exposed over MCP, isolated parallel agents and plugins. Routing is cost aware across 13 model tiers. Each task is scored for complexity and risk by a local Qwen3 classifier and sent to the cheapest model considered capable, from local and low cost models up to Claude Sonnet and Opus for architecture and security review. Routing can step down automatically when a budget cap is near, and failed calls fall back down the chain. Skill protocols are YAML workflows with real gates, so a failing lint, type check or test halts the sequence. Maggy itself is free to self host with the user's own keys. The related Palgu gateway is priced per seat: Free at 0 dollars for one developer and one repository, Team at 39 dollars and Enterprise at 79 dollars per seat per month. Tokens are billed at provider cost plus 10 percent, with no token fee when the user brings their own provider keys.