Live sets
On this page
This guide treats the bridge as an attended performance assistant. It can plan bounded soundscape intent, inspect and shape selected Grid devices, apply revision-guarded graph changes, expose transport/track/mix/device controls, and invoke reviewed native action IDs. The producer still chooses the arrangement and decides when a change is musical.
The source currently registers 57 MCP tools; the running tools/list and
native protocol-4 capabilities responses remain authoritative. This page
does not add arranger, clip, or project-delivery APIs.
The live-set model
A useful live set has three layers:
- Material: kick, bass, percussion, texture, cues, and returns with clear frequency and rhythmic jobs.
- Sections: a human-defined energy arc or scene plan.
- Agency: a small set of bounded gestures with a known neutral value.
Use grid_soundscape_plan for a non-mutating plan from a brief, then resolve
all package/module/port identities against current native reads. A plan is not
a project patch and does not create scenes, clips, or instruments.
Example plan request:
{
"brief": "142 BPM minimal raw techno; stable kick and bass, sparse metallic accents",
"style": "deep-bed",
"density": 0.28,
"motion": 0.38,
"contrast": 0.55,
"temperature": 0.35
}
Use grid_list_soundscape_styles for current styles. Planning output is an
intent; it contains no durable native instance IDs, package UUIDs, or assumed
track indexes.
Start from current state
Before each live pass:
- call
get_grid_capabilities,get_selected_device_state, andgrid_song_state; - call
grid_list_trackswhen track indexes matter; - call
grid_list_devicesfor a target device chain; and - call
get_grid_graphandgrid_analyzeonly when the selected device reports a supported graph.
Record the project token, track index, selection token, mix revision, and graph
ID/revision from those reads. A null selection token can support public reads
but cannot authorize device insertion/removal. A false graph_available is a
hard boundary for graph work, not a reason to guess a module or cable.
Build the material layer
For each role, prefer the smallest operation that produces the intended result:
- use
grid_shape_start/grid_shape_composefor exposed selected-device parameters; - use
search_grid_modulesandsearch_grid_modulatorsfor current package metadata; - use direct guarded graph tools for one insertion, parameter, connection, or movement; and
- use
grid_patch_start/grid_patch_stepfor a multi-operation topology change that needs preview and per-step read-back.
Read the graph immediately before each operation and again afterward. Keep kick and sub paths centered and mono; place width, feedback, and bright nonlinear detail above the useful bass register. These are producer design guidelines, not additional bridge guarantees.
A practical role order is:
| Order | Role | Musical question |
|---|---|---|
| 1 | Kick | Does the downbeat stay stable at the intended level? |
| 2 | Bass | Does the bass interlock without masking the kick? |
| 3 | Percussion/tops | Do rests and accents create groove rather than noise? |
| 4 | Texture | Does motion preserve pulse and tonal center? |
| 5 | Returns/cues | Does the effect mark a boundary without smearing it? |
Design agency as a control contract
For every gesture, record:
| Field | Requirement |
|---|---|
| Destination | One primary musical job, such as bass bite or transition depth. |
| Safe range | The smallest meaningful range that does not lose the role. |
| Neutral | A known value that restores the accepted groove. |
| Time scale | A phrase, bar, or gesture duration. |
| Reset | A clear action or return ramp to neutral. |
Prefer one reviewed modulation route at a time. Use current live parameter metadata and exact graph revisions; historical package names and inventory counts are not substitutes for current catalog reads.
Use host controls safely
For transport, tracks, and mix:
- read
grid_song_stateand (for mix)grid_inspect_mix; - use the fresh project token/mix revision for one guarded write;
- read state again; and
- continue only when the requested value is observed.
grid_song_transport supports play, continue, restart, stop, seek, tempo,
meter, and loop. grid_set_track_state supports arm/mute. grid_set_mix
supports gain/volume, pan, and sends. Track indices are 0..63; transport and
normalized-value bounds are in the tool schema. Recording and automated targets
are guarded.
Native public writes return submitted/immediate proxy observations. A
verified: false response is expected and means a fresh read-back is required,
not that the producer should immediately retry.
For devices, search with search_grid_devices, inspect the bounded bank with
grid_list_devices, and use a fresh non-null selection token for
grid_insert_device or grid_remove_device. Re-list after the operation.
Positions are before, after, start, end, and replace.
For dynamic application gestures, call grid_list_actions with a focused
query, review the exact returned action ID, and invoke it through
grid_invoke_action. An action response proves invocation, not the musical
result. Do not infer IDs from display labels or another Bitwig version.
Recovery and listening
Before a structural or parameter batch, save a parameter snapshot or preserve the current graph/project copy. Match recovery to the narrowest scope:
grid_shape_undofor a completed shaping session;apply_parameter_snapshotfor exposed controls;grid_patch_restorefor a completed parameter-only patch session; orgrid_project_undoonly when the host-wide operation is understood.
After recovery, re-read state. For an ambiguous result, stop and read before retrying. During performance, use the planned neutral values rather than guessing a landing point.
State read-back cannot answer whether the music is compelling. Human listening is separate from bridge acceptance; no native audio capture or rendering tool is available.
Explicit capability boundaries
The bridge does not implement:
- native arranger note, clip, or automation editing;
- MIDI import or voice realization;
- live capture;
- audio rendering/export; or
- project save, new, load, or delivery.
An action returned by the native action catalog does not change these declared boundaries. If a desired live-set step is not represented by current capabilities, stop at that boundary rather than claiming it was created.