tordo

Tordo is an agent-facing Ableton Live control toolkit. It gives an AI agent a stable CLI and JSON-plan contract for inspecting, planning, dry-running, applying, and verifying safe changes inside a Live Set.

Tordo is in developer alpha.

Features

• Inspect tracks, scenes, clips, devices, mixer state, Browser items, and MIDI notes.
• Create tracks, scenes, MIDI clips, and notes through explicit JSON plans.
• Dry-run plans before writing to Ableton Live.
• Resolve names against a fresh snapshot before apply to reduce stale-index mistakes.
• Guard writes with expected track, scene, clip, device, and parameter names.
• Load discovered Browser items such as racks and presets.
• Insert native Live devices and adjust supported mixer and device parameters.
• Export snapshots and note archives for offline analysis and diffing.
• Keep musical judgment human-ear-in-the-loop: Tordo can edit state, but it does not hear audio quality.

Requirements

• macOS
• Ableton Live 12.4 Suite or newer
• Python 3.11+
• uv, pipx, or another Python CLI installer

End-User Installation

Recommended setup is agent-led:

1. Open a skill-capable AI agent, such as Claude, Claude Code, Codex, or another assistant that can install or read agent skills from a public GitHub repo.

2. Paste this request:

   Add the Tordo skill from https://github.com/deadjoe/tordo/tree/main/skills/tordo,
   then set it up for Ableton Live on this Mac and check everything works.
   

3. Let the agent inspect the Skill, install or expose the tordo CLI with your approval, run tordo doctor, install the Ableton Remote Script, and guide you through the required Ableton Live restart and Control Surface selection.

After setup, open a Live Set and ask the agent for the musical change you want. Tordo previews changes before applying them.

Skill support differs by agent. If your agent cannot install a Skill directly from GitHub, clone this repository and point the agent at skills/tordo/.

Manual CLI setup is also available. For normal use, install the released CLI from PyPI:

uv tool install tordo

Install the Ableton Remote Script into your Ableton User Library:

tordo install-remote-script

Restart Ableton Live, then select this control surface:

Settings -> Link, Tempo & MIDI -> Control Surface -> TordoBridge

Input and Output can stay set to None.

Check the setup:

tordo doctor
tordo ping
tordo snapshot
tordo schema
tordo capabilities

Agent Skill

The first agent-facing Skill lives in skills/tordo/.

Give a compatible AI agent this Skill path:

https://github.com/deadjoe/tordo/tree/main/skills/tordo

The Skill teaches the agent to:

• check whether the local tordo CLI is installed
• propose a tordo CLI install command and ask before running it
• run tordo doctor, tordo schema, and tordo capabilities
• install the TordoBridge Remote Script into Ableton Live
• discover installed Packs, User Library, and Current Project Browser resources
• dry-run every plan before apply
• verify changes after apply
• ask for human listening feedback when a decision depends on musical taste

Basic Plan Flow

Plans are JSON documents applied through apply-plan.

Dry-run first:

tordo apply-plan plan.json --prepared-out prepared-dry-run.json

Apply only after inspection:

tordo apply-plan plan.json --apply --prepared-out prepared-apply.json --timeout 120

Example operation:

{
  "plan_version": 1,
  "name": "shape-main-hook",
  "operations": [
    {
      "type": "set_device_parameter",
      "track_name": "Main Hook",
      "device_index": 0,
      "device_name": "Antenna Lead",
      "parameter_index": 6,
      "parameter_name": "Echo",
      "value": 8.0
    }
  ]
}

The CLI resolves names to current indices from a fresh snapshot, then adds expected-name guards before the plan reaches Live.

Safety Boundaries

• Never assume a previously observed track index is still correct.
• Resolve names from a fresh snapshot immediately before write operations.
• Dry-run destructive or large writes first.
• Do not overwrite existing clips implicitly.
• Do not delete tracks or scenes without explicit destructive permission.
• Do not delete the final regular track in a Live Set.
• Treat installed Packs, User Library, and Current Project Browser contents as user-specific resources that must be discovered before use.

Developer Notes

Set up a local checkout:

uv sync --dev --frozen

Run the local quality gates:

uv run ruff check .
uv run python tools/check_operation_registry.py
uv run python -m unittest discover -s tests -p 'test*.py'
uv run python -m py_compile tordo/*.py remote-script/TordoBridge/bridge.py tools/*.py tests/*.py skills/tordo/scripts/*.py

Build and validate package artifacts:

rm -rf dist
uv build
uv run python tools/check_package_artifacts.py
uv publish --dry-run --trusted-publishing never dist/*

The wheel contains the Python CLI plus the packaged TordoBridge Remote Script source. The sdist is intentionally scoped to public release files and excludes non-release development materials, local test material, and runtime artifacts.

Versioning

The Python package version and Live bridge version are intentionally separate:

• The package version is the CLI and Skill distribution version shown by PyPI.
• The bridge version is the Live-side Remote Script compatibility version checked by tordo doctor.

The current bridge source version is TordoBridge 0.8.1.

License

Tordo is licensed under the Apache License, Version 2.0. See LICENSE.

Trademark Notice

Tordo is an independent project and is not affiliated with, authorized, sponsored, or endorsed by Ableton AG. Ableton and Live are trademarks of Ableton AG.