Blog
Read the latest news from the Chatmu blog.
An AI VST plugin generator turns a plain-language description into an instrument or effect you can play. With Chatmu Instrument and Chatmu FX you describe a sound and a look inside Ableton Live, and the Chatmu agent builds a playable plugin on the track in front of you, with no code and no compiling. Both plugins come with Chatmu Desktop, which launched on October 31, 2026.
Updated October 2026.
I'm a producer and sound engineer, and the part of sound design I always wanted to skip was the hunting: scrolling presets and opening five synths to find the one that can do the thing in my head. So we built a plugin that starts from the thing in your head.
They are two blank plugins the agent designs for you. Drag Chatmu Instrument onto a MIDI track or Chatmu FX onto any track, then tell the agent what you want: "a dubstep bass with a pixel-art face", "a dark plate reverb", "a lo-fi box that makes drums sound like an old cassette". The agent writes two things:
The plugin changes live, on the track, while Ableton keeps running. Chatmu is an AI agent for the music industry with its own catalog of music-industry MCP connectors, and Desktop is where that agent works inside your DAW.
Most "make a VST with AI" tools ask a model to write C++ and then compile it. We took a different route. The agent never writes code. It writes a patch: a list of modules and how they connect, plus the controls it wants to expose.
The patch is assembled from 31 DSP modules. Sources include analog-style, wavetable, FM, additive, chip (8-bit), noise, plucked-string and modal (mallets, bells, kalimba) oscillators, synthesized drums, a sampler and granular. Then filters (state-variable, ladder, formant, parametric EQ), modulators (envelopes, LFOs synced to tempo, macros, a step sequencer, an arpeggiator), and effects such as drive, bitcrush, chorus, phaser, delay, algorithmic reverb, compressor, lo-fi, vocoder, pitch shift and multiband processing.
Every patch maps its controls onto 64 generic parameters, so the knobs you see on the face are the same ones you can automate in Ableton. Change the patch and your automation lanes stay in place.
When the agent sends a new patch, the plugin builds it in the background and swaps it in with a short crossfade. Notes that were already playing carry over. You don't reload the plugin, rescan anything or restart Live.
If a patch is invalid, the plugin rejects it with a readable error ("module lfo2 doesn't exist; did you mean lfo_2?"), the agent fixes it, and the previous sound keeps playing in the meantime. If a patch ever produces broken audio (NaN values or a runaway level), the plugin silences that block and goes back to the last good version on its own. A bad idea never crashes your session.
The plugins don't open blank. Each one ships with a factory set you can play right away and use as a starting point for your own ideas.
| Chatmu Instrument (factory) | Chatmu FX (factory) |
|---|---|
| Dream Pad | Tape Box |
| Wobble Machine | Plate Reverb |
| Arcade 8-bit | Glue Comp |
| Pluck & Mallet | Lo-fi Box |
| 808 | Talk Box |
| Drum Kit | Multiband Saturator |
| Lo-fi Keys |
Each factory plugin has its own engine and its own face, and each comes with several sounds inside. Tape Box, for example, is one effect chain with a handful of variations, from subtle to "old VHS". If you edit a factory plugin, your version is saved as a copy in your library and the original stays untouched.
From there, the custom synth plugin AI workflow is just a conversation. A few prompts I use:
The same applies to the AI effect plugin side: "a dark plate reverb with a photo of me in the background", "a sidechain pump that follows my kick". Chatmu FX has a sidechain input you can route in Ableton.
Here is the whole loop, step by step.
Presets work on two levels. A plugin is the face plus the engine ("Wobble Machine"). A sound is a set of values inside that plugin ("Drop", "Intro", "Growl"). Switching sounds is instant and doesn't reload the face.
There are good tools in this space, and they solve different problems. Here's a fair comparison based on their public pages.
| Chatmu Instrument & FX | Pluginmaker.ai | Vibe-coding with JUCE (e.g. Audio Plugin Coder) | ChatDSP | |
|---|---|---|---|---|
| Where you work | Inside Ableton, with Chatmu Desktop | In the browser | Your code editor with an AI coding agent | Inside Ableton, as a Max for Live device |
| What you get | A patch and a face loaded into Chatmu Instrument or FX | A downloadable plugin for DAWs that support VST or AU | A JUCE project with full source code | A new Max for Live device |
| Code or compiling | None | None for the user | You build it (CMake plus Xcode or Visual Studio) | None |
| Hears your mix | Yes, through the Coproducer on your Master | Preview inside the browser | Not part of the workflow | Not described |
| Requirements | Chatmu Desktop and Ableton Live | An account; credit-based | A coding agent, JUCE toolchain | Live with Max for Live, plus your own Anthropic or OpenAI credits |
Where Chatmu is different: nothing gets compiled, nothing becomes code, and the plugin lives inside your session next to the agent that made it. That agent can also see your tracks and hear the mix, so "make this bass sit under the kick" is a request it can check by listening, not a guess. And it's the same agent that can later export the song and take it to distribution, video and social from Chatmu.
This is not AI-generated music. It's AI inside your DAW. You play the notes, you decide, everything stays editable, and it sounds like you.
Chatmu Instrument and Chatmu FX come with Chatmu Desktop, which launched on October 31, 2026 for Mac (Apple silicon and Intel) and Windows. It starts with Ableton Live 11 and 12; more DAWs are coming. Desktop also includes the Coproducer and Chatmu Sound, the synth that can recreate a sound from a recording. There's a free plan; see pricing. If you'd rather use Chatmu's music-industry tools from Claude or ChatGPT, see Chatmu MCP and our complete guide to MCP in the music industry. For the bigger picture of how one agent runs the business side, read AI agent for your music business.
Yes. Some tools generate code you compile; others, like Chatmu Instrument and FX, have the AI write a patch and an interface that load instantly into a plugin already on your track.
No. You describe the sound and the look in plain language. The agent builds the patch from 31 DSP modules and writes the face.
No. Invalid patches are rejected with an error the agent fixes, and if a patch produces broken audio the plugin returns to the last good version automatically.
Yes. Each patch exposes up to 64 parameters that Ableton can automate, and they stay mapped when the patch changes.
Chatmu Desktop starts with Ableton Live 11 and 12. More DAWs are coming.
AI for the industry. Humans for the music.
