Audio tools for your AI agent.

Start with a sound or a file.

• “Make three short metallic kicks and measure their loudness.” Generation needs downloaded weights and a supported runtime.
• “Measure the LUFS and true peak of this WAV.” Analysis needs no model.
• “Load this sample into PAM and build a pattern.” Enable the MCP Bridge under Settings → Integrations in the running PAM instance, then select it through the same server.

$ pam audio generate --prompt "dusty hip hop breakbeat, dry drums, 92 bpm" --seed 7 --duration 5.217391 --category loop --out dusty-breakbeat.wav --json

{  "ok": true,  "sampleRate": 44100,  "seed": 7,  "wavPath": "dusty-breakbeat.wav"}

$ pam audio analyze dusty-breakbeat.wav --json

{  "channels": 2,  "durationSeconds": 5.217392290249433,  "format": "WAV file",  "integratedLufs": -17.51082901429753,  "loudnessRange": 0.9000000000000021,  "ok": true,  "path": "dusty-breakbeat.wav",  "samplePeakDb": -0.9999985473508068,  "sampleRate": 44100,  "truePeakDb": -0.9972471932198774}

{"jobId":"01accb45229d4e79a43d1fc7fa2f92eb","progress":0}
{"jobId":"01accb45229d4e79a43d1fc7fa2f92eb","progress":1}

Connect your agent.

Use a local agent with shell execution or local MCP support. Hosted web and mobile chat cannot execute the binary on your computer. pam mcp gives the agent one pam server, backed by pamd, for files, models and your selected PAM instance. Replace older pam-ai and pam-mcp entries when migrating.

To install the command-line tools, run PAM’s macOS installer from https://map.audio/download, click Customize and tick “Command-Line & AI Agent Tools” (Apple Silicon Macs). Then add PAM to your agent’s MCP configuration:

{
  "mcpServers": {
    "pam": {
      "type": "stdio",
      "command": "/usr/local/bin/pam",
      "args": [
        "mcp"
      ]
    }
  }
}

Inspect the commands available on your machine:

pam version --json
pam manifest --json
pam search "measure loudness"
pam schema audio generate --json
pam audio analyze recording.wav --json

Inspect the resulting file and its measured output, then deliver it through your client’s file tools. A local WAV path alone does not prove an attachment, upload or audition.

How the tools work.

pam

Commands and JSON results for local audio tasks, model management and discovery.

pam mcp

The same command contract as MCP tools, with search, describe and run for discovery.

pamd

The persistent local audio service behind agent workflows: retained model workers, job events, cancellation and retrievable results. Only pam starts it.

PAM instrument

Hosts its own models in-process. Authenticated bridge tools edit the instance selected by each agent session.

Inference and measurements run on the computer. Your chosen AI client handles conversations, tool results and shared attachments.

Compare the workflows.

Competitor capabilities below come from publisher documentation and source; they were not independently tested. No overall superiority claim is established. Checked 2026-10-18.

Free local models. PAM’s engine when you need it.

Local model generation, streaming and audio analysis are free. Use fx, audio master, preset render and dj set to process files and render presets and DJ sets without opening PAM; they require a PAM licence or active trial.

Model terms apply separately. Stable Audio 3 needs about 2.7 GB of weights; its publisher’s commercial licence has registration and revenue conditions. Model inference requires Apple Silicon and downloaded weights; inspect the installed catalogue for runtime availability.

Powered by Stability AI. Read the model licence.