Changelog
All notable changes to Groove AI.
v0.1.0 beta 2026-04-05
added The daemon — the brain of the operation
Groove now has a persistent daemon running on :3141 that manages all agent coordination. It holds the registry, broadcasts state over WebSocket, and exposes a REST API. This is the foundation everything else builds on — agents can finally be aware of each other because there's a central process keeping track. State persists to disk, so if the daemon restarts, your agents don't vanish.
added Spawn, kill, and manage agents from the CLI
The core commands are in: groove start, groove stop, groove spawn, groove kill, groove agents, and groove nuke. The goal was to make multi-agent workflows feel as simple as opening a new terminal tab. Spawn an agent with a role and scope, and it's immediately introduced to its peers. No config files, no ceremony.
added Introduction protocol — agents that know about each other
This is the thing that makes multi-agent actually work. When you spawn agent #2, the daemon writes AGENTS_REGISTRY.md and notifies all existing agents about the new arrival. Every agent knows who else is running, what files they own, and what they're working on. No more two agents silently editing the same file and creating contradictory changes.
added File lock manager
Scoped file ownership is now enforced at the daemon level. When you spawn with --scope "src/api/**", that agent owns those files. Other agents are told to stay out. We went with advisory locks for v0.1 — agents are informed, not blocked. This prevents the #1 pain point in multi-agent setups: wasted tokens from conflicting edits.
added The Journalist — first pass
The Journalist is now running as a background synthesis loop. It tails agent stdout, filters noise, and generates GROOVE_PROJECT_MAP.md — a living snapshot of what exists, what's being built, and who's doing what. This is the cold-start killer. When a new agent spawns, it reads one file and knows the entire project. No more wasting 5 minutes and thousands of tokens re-explaining your codebase.
added Claude Code as the first provider
Day 1 provider support ships with Claude Code. The provider abstraction is in place — isInstalled(), buildSpawnCommand(), parseOutput() — so adding Codex, Gemini, and Ollama is just implementing the interface. But Claude Code is where most of our users live, so that's where we started.
v0.0.1 2026-04-04
added Project scaffolding
The monorepo is set up with npm workspaces — packages/daemon, packages/cli, and packages/gui. Added CLAUDE.md so AI coding agents working on Groove itself have full context on the architecture. Licensed under FSL-1.1-Apache-2.0 — source-available now, converts to Apache 2.0 after two years. We want to build in the open without giving away the farm on day one.