Agent-Aware CLI
Dual-mode terminal execution. Human developer ergonomics by default; passing --json guarantees deterministic, single-object JSON on stdout for automated agents.
Audio Tools for Agents
Deterministic audio manipulation and local deep-learning stem separation for AI agents.
Core Delivery Surfaces
Section titled “Core Delivery Surfaces”Model Context Protocol (MCP)
Standard stdio JSON-RPC server implementation. Allows Claude, OpenCode, and Cursor agents to inspect audio and isolate stems without shell subprocess management.
Agent Skills Spec
Progressive disclosure skill package complying with agentskills.io. Includes discovery frontmatter, model selection rubrics, and non-interactive runner scripts.
Meta HTDemucs Neural DSP
Official Meta AI PyTorch inference engine. Separates audio into 4 stems (vocals, drums, bass, other) or 6 stems (+ guitar, piano) with live memory telemetry.