ai-devkit

Why we included this project

When a team outgrows a single assistant and finds itself running Claude Code, Codex, Gemini CLI, Cursor, and opencode in separate terminals, the hard part stops being any one tool and becomes the lack of a shared way to control them all. AI DevKit sits on top of those agents as a local-first layer. One .ai-devkit.json reconciles setup across tools, an agent console TUI shows live sessions, and agent send routes prompts and test output into a running agent. A local SQLite memory store keeps decisions and conventions searchable instead of stuffing them into every prompt, and composable skills like the dev-lifecycle workflow plus a verify gate let teams build a repeatable, reviewable operating model that won't call work done without fresh test evidence. That makes it most useful for engineering leads and solo maintainers whose AI setups have sprawled into a handful of isolated terminals. It isn't another coding model; it's the glue between the ones you already run.