Blokkily demos a music workstation that agents can edit
The experimental DAW combines a human-approved LLM pattern assistant with an MCP bridge that exposes the running song to agents.
Two ways to bring AI into a song
Blokkily’s author presented an experimental, pattern-first digital audio workstation in an October 12 Show HN post. Its tracker, piano roll and step grid edit the same underlying musical events. The project README describes an LLM assistant that turns prompts such as an ascending scale or offbeat drums into proposed notes. A producer must press Apply before the proposal changes the song, and the resulting edit can be undone. The documentation says the assistant can use a local Ollama model by default or configured remote endpoints. Those are described features of the project, not an independent test of musical quality.
The workstation is also an agent tool
Blokkily exposes an MCP bridge that can read a session, write notes, control playback, select instruments and capture a screenshot of the running window. The author says agent edits use the same undoable paths as edits made in the interface. That makes the Show HN submission a concrete example of an AI agent operating creative software, beyond generating an audio file from a prompt. The repository dates to September and offers source-code build instructions; the October 12 event is the public demo, not evidence of a new packaged release or broad adoption. The README labels the workstation experimental and lists unfinished features.