Catalog
Persistent local index of Serum presets, samples, grooves, and device presets, with similarity search and reuse-vs-create recommendation that can orchestrate the existing create tools.
Tools on this page
search_catalog
search_catalog(
query: str,
domains: list = None,
kinds: list = None,
sources: list = None,
limit: int = 50,
)
Search the local content catalog (Serum presets, samples, grooves, device presets) by name/folder/tags. Fast persistent index; auto-refreshes changed filesystem roots. Returns {ok, data:{count, returned, truncated, items}}.
find_similar
find_similar(intent: str, domain: str, limit: int = 10)
Rank catalog assets in one domain by lexical similarity to an intent string, with scores. domain in serum|sample|groove|device_preset. Returns {ok, data:{items:[{name, path, uri, source, folder, kind, score}]}}.
catalog_stats
catalog_stats()
Counts and freshness for the local catalog by domain/source/kind, plus per-root last-indexed times. Returns {ok, data:{total, by_domain, by_source, by_kind, roots, device_presets_last_indexed}}.
reindex_catalog
reindex_catalog(domains: list = None, full: bool = False)
Rebuild the catalog. domains defaults to all. full=True drops and rebuilds from scratch. Device presets reindex only when Live is connected. Returns {ok, data:{indexed:{domain:count}, errors:[…]}}.
recommend_asset
recommend_asset(intent: str, domain: str)
Propose reuse-vs-create for a described asset. domain in serum|sample|groove|device_preset|wavetable|simpler. Returns {ok, data: {decision, top_matches, create_plan, reason}}. PROPOSE ONLY - never writes; pass create_plan to create_from_plan(confirm=True) to act on a ‘create’.
recommend_instrument
recommend_instrument(intent: str, prefer_native: bool = None)
Choose which instrument engine to reach for from a free-text intent (e.g. ‘gritty evolving reese bass for dnb’). Ranks Serum vs native Wavetable, Operator, Analog, Simpler/Sampler, and Drum Rack against a capability matrix, biased toward native engines so Serum is used only when it earns it. Applies live signals when connected: Serum-installed (drops Serum if absent), Serum instance count (busy penalty), and portability (prefer_native). Set prefer_native=True to force native-only weighting; leave None to auto-detect (native-preferred only when the set has no 3rd-party plugin). PROPOSE ONLY - pass the returned top.domain to recommend_asset to pick reuse-vs-create. Returns {ok, data:{intent, parsed, ranked, top, reason, environment, next}}.
create_from_plan
create_from_plan(plan: dict, confirm: bool = False)
Execute a create_plan from recommend_asset. confirm=False (default) returns a dry-run describing what would be created and writes nothing. confirm=True invokes the matching creator (Wavetable, Simpler, or a Serum base-preset clone). Returns {ok, data:{…}}.
recommend_track_instrument
recommend_track_instrument(
track_index: int,
clip_index: int = 0,
clip_location: str = 'auto',
intent: str = '',
limit: int = 5,
include_plugins: bool = True,
)
Classify a MIDI track’s part from its notes and rank presets for it.
Measures the clip (pitch range, polyphony, note length, density), classifies the musical texture (bass / pad / keys / stabs / lead / arp / drums …), uses the track name as a prior (content wins on a conflict, which is reported), then ranks loadable stock presets from the catalog for that role and lists installed plugins that suit it. intent adds free text to the ranking (“warm”, “boom bap”, “orchestral”). clip_location “auto” reads session slot clip_index and falls back to arrangement clip clip_index. PROPOSE ONLY - hand chosen {track_index, uri} pairs to apply_instrument_plan. A plugin candidate loads bare (patch_selection: manual - the LOM cannot pick a Kontakt library or a VST patch). Returns {ok, data:{track_index, name, clip, analysis, intent, current_devices, doubles_audio_track, candidates, plugins, next}}.
recommend_set_instruments
recommend_set_instruments(
track_indices: list = None,
intent: str = '',
limit: int = 3,
include_plugins: bool = True,
)
recommend_track_instrument for every MIDI track, with set-wide context.
Covers every MIDI track that holds a clip (or the given track_indices), in one call with set-wide context: tempo, a key guess from all parts together, instruments already in use (a sound used on another track is ranked down so parts do not collide), and whether a part doubles an audio track of the same name. Returns {ok, data:{tempo, key_guess, parts:[…], plan:[{track_index, name, uri, candidate}], skipped, next}}. plan holds each part’s top stock candidate, ready for apply_instrument_plan after review. PROPOSE ONLY.
apply_instrument_plan
apply_instrument_plan(assignments: list, confirm: bool = False)
Load a reviewed instrument plan (hot-swap) and verify each device.
assignments is a list of {track_index, uri, candidate?} (the plan recommend_set_instruments returns, edited as needed). confirm=False (default) is a dry run that writes nothing. confirm=True loads each uri (HOT-SWAP: the track’s current instrument is replaced) and verifies the device landed by re-reading the track. Returns {ok, data:{applied:[{track_index, name, uri, devices_after, verified}], failed:[…], dry_run}}; ok is false when any track failed.