Ismail Builds a Text-Operated DAW for AI Agents

Summary

Ismail is an MIT-licensed digital audio workstation designed to be operated by AI agents rather than by a person using ears and a mouse. It is not a prompt-to-song model like Suno; an agent writes notes, instrument patches, effects, automation, and code, renders the project, reads structured audio analyses, and edits the result. The project exposes the same operations through an MCP server with about 70 tools, a command-line interface, and a Python API, with an agent skill that recommends planning a Session Sheet and writing a Listening Report after each render. Its analysis tools convert rendered audio into text describing grids, levels, spectra, chords, melodies, drum patterns, piano rolls, timbre, formants, arrangement, and comparisons with reference tracks. The engine includes synthesizers, samplers, drum instruments, code-defined voices, effects, automation, undo snapshots, caching, and batch operations. Projects are folders containing code and data, so Git history, diffs, branches, and code review can be used for tracks. Ismail runs locally on Windows, macOS, and Linux, requires Python 3.10 or newer, and optionally supports perceptual metrics, stem separation, and music-video generation. The repository says stronger models can improve the agent’s musical output, while acknowledging that systems such as Suno remain better at realistic sung vocals and fast polished results.