Bionic (LM Studio)

Use Producer Pal completely offline without an Internet connection.

Bionic is LM Studio's agent app for open models. It bundles the LM Studio runtime, so it downloads and runs local models on its own — you don't need the classic LM Studio app installed alongside it.

Bionic vs. classic LM Studio

These are two separate apps from the same team, and LM Studio still ships both. Producer Pal's offline setup guide targets Bionic, which is where new work is going. The classic LM Studio app still works with Producer Pal, but it's no longer officially supported here.

Experimental

This requires a relatively new machine with decent specs (Apple Silicon with lots of RAM or PCs with Nvidia 4080+ graphics cards). It requires more technical know-how to setup and debug. The online options work significantly better and faster at the time of writing. However, completely offline and private usage is compelling.

Requirements

Download Max for Live DeviceProducer_Pal.amxd — add it to a MIDI track in Ableton Live

• Ableton Live 12.3+ with Max for Live. Live 12.4 or later is recommended — some features don't work on older versions of Live. Use the version of Max bundled with Live, or make sure your standalone Max is up to date.

• Bionic

Bionic is free to download, and running models locally is free. An LM Studio account and billing are only needed for cloud models and web search, neither of which Producer Pal requires.

Installation Steps

1. Install the Max for Live Device

Download Producer_Pal.amxd, the Producer Pal Max for Live device, and add it to a MIDI track in Ableton Live:

It should display "Producer Pal Running" or something isn't working.

2. Install a Compatible Model

Download a model in Bionic under Settings → Explore, such as a recent version of:

• Qwen
• Gemma
• Mistral Magistral or Devstral
• GLM Flash

It has to support tools. Your downloaded models are listed under Settings → Library, where a hammer icon marks the ones that do:

3. Add Producer Pal to Bionic

Go to Settings → Connected Apps and click "+ Add custom MCP" under Manual MCP server setup:

Option A: With npx (recommended) — Allows flexible startup order and auto-reconnection:

• Name: Producer Pal
• Connection: On this computer
• Command: npx
• Turn on Show advanced options, then use + Add argument to add -y and producer-pal as two separate arguments

The producer-pal package is a proxy that responds to requests even when Ableton Live or the Producer Pal device are not running, to let you know there's a problem.

Option B: Direct HTTP — Requires Ableton running first, no auto-reconnection. Set Connection to Web address and use the URL http://localhost:3350/mcp.

Click Add MCP. Producer Pal should show as connected, with its tools ready:

Restart Bionic after changing MCP settings

Adding a server or editing its arguments may not take effect until you restart the app.

4. Tune It for Your Model (Recommended)

Bionic has no tool picker, so the MCP server's arguments are the only place to narrow what Producer Pal offers — and small models do much better with a short tool list. The same arguments also switch on the settings that make Producer Pal easier for a local model to drive:

• --small-model-mode — a smaller, simpler interface built for local models
• --notation stark — a literal type: content format with event-based drum hits, which small models handle better than the default bar|beat text
• --tools core,clip,track — a focused toolset, which also shrinks the skills the model reads on connect

Edit the server under Settings → Connected Apps and add them as arguments, one field each — a flag and its value are two separate arguments:

--tools keeps only what you list; --disable-tools drops what you list. Both take tool names (read-clip or ppal-read-clip) and group names: core, session, actions, live-set, track, scene, clip, device, advanced, and read-only. Run npx producer-pal --list-tools to print the groups plus the tools the running device currently offers.

Withholding a tool also drops the part of the Producer Pal Skills that teaches it, so you stop paying for the tool's schema and its guidance in every conversation. --tools read-only cuts the skills text by more than half.

Unlike the other flags, this one is per client: the Chat UI and your other MCP clients keep the full toolset. ppal-connect is always kept — it is how the AI connects and receives the skills.

One wrinkle with --tools: it keeps what you list by withholding everything else, and "everything else" is the tool list this copy of npx producer-pal knows — so a tool added in a newer Producer Pal stays enabled until you update. --disable-tools names tools directly, so it can withhold a newer tool even from an older copy; only its group names are limited to the ones above.

Small model mode and notation apply to the device itself, so they also change what the built-in chat UI and any other connected client see. You can set them by hand instead on Producer Pal's "Setup" tab:

5. Start Using Producer Pal

Bionic works in a project tied to a local folder, so create one with New Project before starting a conversation. An empty folder is recommended for Producer Pal sessions. Feel free to put reference documents in it, such as common workflow instructions or details of your preferred musical style and production techniques.

Start a conversation and say "connect to ableton":

Local Model Tips

• A context length of about 8k (8000) is needed to start a conversation with Producer Pal and send a few messages.
• A context length of 16k (16000) or higher is recommended for achieving useful results, but higher context lengths can make the model run significantly slower, especially as the conversation grows. Experiment to find the right balance.
• Follow-up turns are often faster. The LM Studio runtime automatically reuses its cached prompt prefix between turns, so later messages in a conversation can be much quicker than the first. There's nothing to configure, and the size of the benefit varies by model.
• Add tools back to --tools when you miss features. Every tool you add costs context on every conversation.
• Experiment with empty or extremely simple Live projects
• Only work with very simple material (e.g. basic MIDI patterns in clips 4 bars or shorter)
• Note that small models are guided to delete and start over rather make edits (other than simple additions)
• If the AI struggles and makes mistakes, don't hesitate to delete recent messages from the chat, edit your last message, and try again. Don't waste tokens correcting the LLM. Back up and avoid the issue or try something else.
• Shut down every other app you don't need to be running
• Consider running Ableton Live on a different machine on the local network
• When using a model with the GGUF engine, try enabling the advanced/experimental settings for Flash Attention and setting the K/V caches' quantization to Q8 or Q4.
• Research how to optimize for your specific machine / GPU hardware

Advanced: Customizing Skills

Local models vary widely in capabilities. First, try small model mode (see installation step 4 above) — it uses a simplified instruction set designed for smaller models.

If Small Model Mode isn't working well for you, or you want to experiment further, you can customize Producer Pal's behavior for your specific model:

1. Download this repository and follow the dev setup and build instructions in DEVELOPERS.md
2. Edit the skills files in src/skills/ - Small Model Mode uses the basic driver (src/skills/drivers.ts) with the basic notation head for the active notation (e.g. src/skills/notation/barbeat-basic.ts), while regular mode assembles the standard driver's fragments (src/skills/fragments/) with a standard notation head (e.g. src/skills/notation/barbeat-standard.ts). See src/skills/build-skills.ts for how they are selected
3. Experiment with instruction wording, remove features your model struggles with, or adjust the guidance
4. Rebuild with npm run build
5. Use the development version of Producer_Pal.amxd in Ableton Live
6. Reload your Producer Pal MCP server in Bionic and start a new conversation

Share your findings: If you discover configurations that work well for specific models, please share in GitHub Discussions. The community benefits from learning what works with different local models.

Alternative: Built-in Chat UI

You can also point Producer Pal's built-in chat UI at Bionic's local server and use that interface instead of Bionic's own.

Troubleshooting

If it doesn't work, see the Troubleshooting Guide.