Start here

On this page

Pocket guides#

Every guide here stands on its own, but they read best in roughly this order. If you're new, start with the first two.

Start here#

  • Your first experiment: install Pocket and run a complete example in a few minutes, with no DAW, model or account.
  • Key ideas: stores, receipts, clocks, occurrences and the kinds of evidence, in plain terms, plus a glossary.
  • Use Pocket with an agent: connect an AI assistant over MCP and give it the playbook.

Practice and listening#

These work anywhere Python runs. No DAW is needed.

  • Musical context and practice: mark passages, clocks and cues in a recording, render exact alternatives, compare them and keep attributed reports.
  • Practice review: a local browser page for listening to a comparison and saving what you heard, pinned to the exact audio and interval.
  • Peek: look inside a recording. See where sounds start, which pulse rates fit, how the tempo drifts and which pitch classes ring, without anything being decided for you.

MIDI and sound#

  • MIDI and sound tools: import, generate, edit, structure and export MIDI material with explicit timing, locks and expression. Also covers audio hypotheses, control curves and sound plans.
  • What's supported: what works on its own, what works with Ableton Live, and what's still on the roadmap.

Ableton Live#

Optional adapters for Ableton Live 12. Each guide lists its setup.

  • Thread: read a saved set. See its clips, their sources and where each moment of the arrangement comes from.
  • Stitch: compare exact renders of a transition, and prepare a single controlled change as a separate Live project.
  • Listening notes: find earlier notes about a Stitch trial, scoped to the exact audio and span.
  • Baste: take a read-only look at the session that's open right now.
  • Pipette: carry one explicitly kept trial into a new project, with the original left untouched.

Records and sets#

Tools for planning and playing sets of records. They propose; you decide.

  • Choosing records: how the selection tools fit together, with command-line and MCP examples.
  • Record bags: a sealed crate of records with honest, attributed notes.
  • Weave: sketch several possible orders for a set.
  • Whisker: propose what could come next, in a session you and an agent share.
  • Local workspace: a small browser page for bags, routes and next-record choices.

Optional connections#

Each of these needs its own setup and is never used unless you ask for it.

  • Recording acquisition: fetch one explicitly chosen, authorized recording URL with strict checks.
  • Spotify: import a playlist's catalog, or create a new private playlist from a reviewed plan.
  • Music embeddings: an optional local model for coarse "sounds like" retrieval.

Reference#

  • Provider reference: every callable provider, with its Python signature, CLI command and MCP input schema, generated from the installed package.
  • Contracts and errors: export machine-readable contracts, choose machine errors and retry safely.
  • Architecture: how the library is organized and why.
  • Name changes: earlier tool names and the aliases that still work.
  • Agent skill: the working guide Pocket gives to AI agents.

Project#

Runnable examples live in examples/. They generate their own test material or take a recording you supply.

Pages named on-deck, set-map, set-workshop, track-map and transition-lab are short redirects kept for old links. See name changes.

Improve this page on GitHub