magda-core, from Conceptual Machines, is the engine behind a free open-source DAW designed for AI-agent control of music creation. It offers natural-language DAW control, local model support for offline workflows, and a professional audio engine while keeping a lightweight footprint. Key elements include hybrid audio/MIDI tracks, a clip-based Session View and timeline Arrangement View, aimed at producers, developers, and researchers exploring agentic composition.

What tasks can you actually use magda for?

magda targets recording, editing, arranging, live clip performance, and iterative sound design while exposing session state to external agents. It supports hybrid tracks that host both audio and MIDI, a clip-based Session View alongside a timeline Arrangement View, and third-party plugins such as VST3 and AU. The included Faust interpreter lets users write and test DSP effects in real time, which suits rapid prototyping of custom processing.

How reliable are the tool's generated musical outputs?

The integrated AI console generates MIDI and session edits using the active session context, and structured control via the MAGDA DSL reduces ambiguous agent commands. Output fidelity depends on the processing model you choose because the app supports cloud providers like OpenAI and Anthropic as well as local models such as QWEN. Expect variation in musical detail tied to model selection and prompt specificity, so post-generation editing is commonly required.

Does the app protect session data and fit local-first workflows?

magda can run as a standalone DAW or act as an MCP server for external agents, and it explicitly supports local models to enable offline operation and a local-first privacy posture. Cloud provider integration is optional. The JUCE and Tracktion Engine foundation gives developers extension points for automation, while MCP formalizes how agents read and modify session state during scripted or programmatic workflows.

Suitable for automation-minded creators who accept a technical learning curve

magda is a practical option for producers, developers, and researchers who prioritize agent-driven workflows and programmable session control, because it was designed around agent interaction and scriptable session access. Users who prefer conventional point-and-click DAW workflows may encounter a learning curve tied to agent protocols and structured control approaches, so expect an initial investment in understanding how agents map to musical tasks.